Campaigns | Digital Watch Observatory

Weekly #256 UN kicks off Global Mechanism on ICT security, road ahead murky

27 March – 3 April 2026

HIGHLIGHT OF THE WEEK

UN kicks off Global Mechanism on ICT security, road ahead murky

After almost three decades of stop-start cybersecurity negotiations at the UN, the long-anticipated Global Mechanism on ICT security has finally kicked off.

It is the first permanent forum of its kind since discussions on ICT security began back in 1998, and its mere existence says a lot about how far those talks have come.

But if the launch felt like a breakthrough, the organisational session quickly brought things back down to earth. Beyond what was already sketched out in Annex C and the OEWG’s Final Report, it remained unclear how the mechanism would actually function in practice.

When the member states agreed to establish the Global Mechanism in July 2025, they also envisioned that the mechanism would meet in plenary and have dedicated thematic groups (DTGs). These groups are intended to enable more in-depth discussions and build on the outcomes of the plenary. The practicalities of how the dedicated thematic groups should be set up and administered were hotly contested at the organisational session, as they will influence what gets on the agenda, who drives it, and whether this new system can deliver real outcomes over time. No decision was made in any of these matters.

A long-standing point of contention and possibly the most politically-charged was the role of non-governmental actors in the groups. Is it a possibility rather than a standard feature? Does inviting external briefers require member-state agreement on a case-by-case basis? How this is resolved will directly determine the degree of access the private sector, technical community, and civil society organisations have to the DTG process in practice.

The mechanism inherited many unresolved substantive debates from its predecessors. On international law, there is widespread agreement that considerable work remains to be done, but little agreement on how to carry it out. A broad majority of states expressed support for ensuring that the mechanism remains action-oriented, with a strong focus on practicality and the implementation of agreed frameworks on international law, norms, CBMs, and capacity-building. Many delegations, primarily from developing countries, urged the Global Mechanism to prioritise the operationalisation of the UN Voluntary Fund, which was tabled but left unresolved by the OEWG.

Furniture, Table, Dining Table, Architecture, Building, Dining Room, Indoors, Room, Chair, Tabletop

What now? The session closed without resolution on any of its most consequential questions. The Chair will convene informal intersessional consultations to resolve outstanding issues before July, when the mechanism will hold its first substantive session, during which it is expected to discuss substantive matters.

We’ll be monitoring the process closely on our dedicated Digital Watch Observatory web page.epresent isolated outcomes or the beginning of a broader legal shift.

IN OTHER NEWS LAST WEEK

Anthropic scores a temporary win against the US government

A California judge has temporarily blocked the US government from enforcing the ‘supply chain risk’ designation against Anthropic, finding that the company’s actions do not meet the legal definition under Section 3252 of the Title 10 United States Code.

That law defines a supply chain risk as the potential for an adversary to sabotage, maliciously interfere with, or subvert a covered system—covert acts, not public or negotiated positions. The court rejected the notion that questioning or resisting contract terms automatically makes a vendor an adversary.

The ruling emphasises procedural requirements: even when Section 3252 allows bypassing standard debarment processes, the government must document risk assessments, consult relevant agencies, and consider less restrictive alternatives. The court found these safeguards were likely ignored in Anthropic’s case.

Additionally, the judge noted that the designation appeared to be influenced by Anthropic’s public statements and its refusal to support certain government AI uses, raising First Amendment concerns. Anthropic was likely denied due process, receiving neither adequate notice nor a meaningful chance to respond before facing substantial economic and reputational consequences.

The court also found that Anthropic demonstrated irreparable harm, including immediate loss of contracts, damaged business relationships, and reputational impact, supporting the temporary block on the government’s enforcement of the designation.

The Court questioned whether the scope of the government’s special national security authorities was appropriate in these circumstances, emphasising that such powers are generally intended for clear and serious risks.

The verdict, for now. Anthropic’s request for a preliminary injunction in this lawsuit against the administration was granted. The injunction does not resolve the case on the merits; it temporarily stops the contested measures.

What happens next? The administration has appealed, and the appellate court, the US Court of Appeals for the Ninth Circuit, will ultimately decide on the matter. However, a final verdict in this case could be months away.

Meanwhile, the government is also facing another lawsuit from Anthropic, filed in Washington, D.C. In that case, the company is challenging its supply chain designation before a three-judge panel at the D.C. Circuit Court of Appeals, specifically contesting the legal authority invoked under the Federal Acquisition Supply Chain Security Act (FASCA).

Why does it matter? The case highlights broader issues regarding the limits of federal power over private technology companies and the protection of constitutional rights, with potential implications for future government interactions with the tech industry.

Iran issues warning to major US tech firms

Iran’s Revolutionary Guard has issued a statement warning that major US technology companies, including Apple, Google, Meta, Intel, Oracle and Nvidia, could face retaliatory action if further Iranian leaders are killed in targeted assassinations.

‘These companies, starting from 8:00 pm (1630 GMT) Tehran time on Wednesday, April 1, should expect the destruction of their relevant units in exchange for every assassination in Iran.’

The group alleges that these firms are the ‘main element in designing and tracking assassination targets.’

Iran also claimed to have conducted drone strikes against communications and industrial sites in Israel.

Reining in social media for minors as trust in platforms erodes

The first country to introduce a social media ban for minors is now assessing how that ban is working. Early results released by Australia’s eSafety Commissioner show significant action by platforms to prevent users under 16 from holding accounts, but also ongoing challenges in fully enforcing the restrictions. By mid-December 2025, around 4.7 million accounts were removed or restricted, with more than 300,000 additional accounts blocked by March 2026. Despite these reductions, many children continue to retain accounts, create new ones, or pass age assurance checks. Regulators identified several compliance concerns, including platforms that allow repeated attempts at age verification and encourage some users to update their ages. Reporting systems for underage accounts were often difficult to access, particularly for parents.

Indonesia is also checking on progress: its social media restrictions for under-16s went into effect last week, and already Meta and Google have been found non-compliant. Indonesia’s Communication and Digital Minister noted that the two companies were summoned on Monday to undergo checks. Failure to implement the curbs, the ministry has noted, may result in sanctions or even a block on the platform in the country.

Australia’s social media ban for minors has inspired many countries to follow suit. One of them is France, which is moving toward restricting social media use for children under 15, as its Senate approved a plan that differs from an earlier, stricter version passed by the National Assembly. While the National Assembly has backed a strict approach, requiring platforms to delete existing accounts and block new under-15 users, the Senate has proposed a more flexible, two-tier system that would limit only harmful platforms and allow access to others with parental consent. The two versions must now be reconciled, meaning the final shape of the law remains uncertain. Key questions—particularly around how age verification will work—are still unresolved and tied to ongoing EU-level discussions, pushing any real implementation to at least 2027.

The MP that introduced the bill warned that this is a matter of public health, noting that ‘When similar questions arose with products like alcohol or tobacco, we collectively chose to prohibit them, because we considered them public health issues.’

Austria’s government also announced plans to ban under-14s from using social media. The government plans to present a draft law by the end of June. ‘We will no longer stand by as these platforms make our children addicted and, in many cases, ill,’ the Vice Chancellor noted.

After a US jury found Meta and YouTube liable in a social media addiction case, the concept of ‘social media addiction’ is likely to gain more legal and policy traction. In Italy, senators have introduced a draft law that directly targets the role of platforms in shaping user behaviour, proposing limits on default profiling and greater transparency around how algorithms curate content. Backed by the opposition Democratic Party, the proposal shifts responsibility toward platform design itself, arguing that recommendation systems are not neutral tools but deliberate corporate choices with real-world consequences.

It’s therefore unsurprising that a new survey in Switzerland revealed a widespread mistrust of big tech, with a large majority of respondents viewing these companies as primarily profit-driven. Concerns range from the impact on children and growing dependence on foreign tech firms to fears about the broader effects of digitalisation on democracy.

Deadlock at WTO: Moratorium lapse meets plurilateral momentum

At the 14th Ministerial Conference of the World Trade Organization (MC14) in Yaoundé, Cameroon, digital trade dominated the agenda through two parallel tracks—each pointing in a different direction and illustrating both the limits and evolution of the multilateral system.

The moratorium on customs duties on electronic transmissions. The long-standing moratorium—renewed every two years since 1998—expired on 31 March after members failed to reach consensus on the length of a new extension, with differing views among members preventing a deal.

Some members, including the USA, pushed for a longer-term solution, while others, led during the talks by Brazil, favoured shorter renewals to preserve regulatory flexibility in light of rapid technological change, including AI and 3D printing.

In parallel, however, a different dynamic unfolded. A coalition of 66 WTO members announced they would move forward with implementing the plurilateral Agreement on Electronic Commerce concluded in 2024 by the Joint Statement Initiative on e-commerce (JSI), through interim arrangements.

Why does it matter? The lapse does not automatically trigger tariffs, but it creates policy space for countries to impose them. The outcome also meant that a broader set of discussions on WTO reform, which had been politically linked to the approval of the moratorium, remained unresolved.

What’s next for e-commerce discussions? Discussions on the moratorium, the WTO reform, and the future of the Work Programme on e-commerce (WPEC) are expected to continue at the next General Council meeting in May in Geneva. In the meantime, JSI members will continue to seek inclusion of the Agreement under the WTO legal architecture.

For a deeper understanding of MC14 outcomes and implications, join the 14 April webinar ‘WTO deadlock, AI boom: Unpacking MC14 and looking ahead’ co-organised by Diplo, the Digital Trade and Data Governance Hub, and the Geneva Internet Platform. Registrations for the event are open.

China launches World Data Organization

The World Data Organization (WDO) was formally established in Beijing, presenting itself as the first international, non-governmental platform dedicated specifically to global data development and governance.

Conceived as a multistakeholder forum, the organisation aims to facilitate dialogue, rule-making, and cooperation, with a stated focus on bridging the global data divide, unlocking the value of data, and supporting the digital economy.

Its inaugural assembly adopted the organisation’s charter, appointed leadership, and set out priorities around capacity building, regulatory exchange, and technological collaboration.

Why does it matter? There is currently no single global body exclusively dedicated to data governance as a whole—covering economic value, governance rules, development, security, and cross-border flows in an integrated manner. The emergence of the World Data Organization (WDO) is significant because it seeks to occupy that space, positioning itself as a dedicated platform for data governance coordination.

At the same time, it reflects broader geopolitical dynamics: China is not only participating in rule-making but actively building platforms that could influence how digital governance evolves, particularly for developing countries seeking alternatives or complements to existing frameworks.

View all updates

LAST WEEK IN GENEVA

machine, Wheel, Spoke, City, Art, Bulldozer, Fun, Drawing

Last Monday and Tuesday (30 and 31 March), ITU held a two-day workshop on ‘Trustable and Interoperable Digital Identities for Human and Agentic AI’ in Geneva. It brought together stakeholders from governments, industry, academia, and standards bodies to examine technical approaches related to trust frameworks, trust management, security, and interoperability; and to investigate actionable recommendations and consolidated insights to advance standardisation work in the field.

The 2026 Global Digital Economy Conference held its Geneva branch event on Tuesday (31 March), gathering political leaders, business executives, and academics to discuss the development of the global digital economy under the theme ‘Digital Intelligence Without Boundaries: Friendship and Win-Win.’ The event featured high-level dialogues, the launch of the Geneva Office of the Global Digital Economy City Alliance, industry insights on China-Europe cooperation, and targeted networking to foster partnerships.

The Inter-Parliamentary Union (IPU) hosted a webinar on ‘Building AI Literacy in Parliaments‘ on Wednesday, (1 April), to explore how parliaments can develop training and resources to support AI literacy among members, parliamentary staff, and IT teams. The webinar highlighted the IPU Guidelines for AI in parliaments, emphasising that AI literacy should reach all roles within parliaments.

To prepare for the 2027 Geneva AI Summit, the Swiss Government invited ICT4Peace to organise and host a launch event at GenAI Zürich yesterday (2 April). The event brought together 40 participants from government, business, academia, and civil society to begin shaping the Summit’s objectives and exploring potential concrete outcomes. Participants discussed a set of guiding questions to shape the focus and outcomes of the 2027 summit. These included identifying areas where international dialogue and cooperation are needed, defining potential political and practical outcomes, and exploring Switzerland’s strengths in facilitating multistakeholder engagement. The discussions also addressed identifying potential partners, resolving areas of disagreement around specific policy objectives, and developing concrete tools and solutions to present as Swiss contributions at the summit.

READING CORNER

The role of the World Data Organization on the digital sovereignty chessboard – Diplo

Beijing hosted the founding assembly of the first international organisation dedicated specifically to data governance and development.

Statecraft in the age of the algorithm – Diplo

As diplomacy migrates from the deliberate silence of morning cables to the relentless vertical scroll of TikTok, explore the new privatisation of statecraft.

Digital Watch newsletter – Issue 108 – March 2026

March 2026 in retrospect

In our March 2026 Monthly newsletter, we observed the deadlock at WTO MC14 over the WTO e-commerce moratorium which led to its lapse, even as a coalition advanced a plurilateral digital trade deal. We examined what it means and what comes next.

Two recent US jury verdicts found Meta and YouTube liable for harms to minors, including exposure to sexual content and social media addiction. Taken together, the cases move beyond questions of content moderation and into the design of the platforms themselves.

The long-awaited Global Mechanism has finally launched, creating the UN’s first permanent forum on ICT security since 1998—but its inaugural session left many questions about how it will actually work. Here’s why it matters and what to watch as the Mechanism takes shape.

Will we one day buy intelligence on a meter? It is not yet certain, but the mere idea raises questions about control, access, and how we measure and consume intelligence in the future.

Plus: March’s top digital policy developments and a Geneva wrap-up.

Snapshot: The developments that made waves in March

Technologies

A new five-year development plan approved by lawmakers in Beijing centres on innovation and advanced technology to drive future economic growth and global leadership, prioritising AI, robotics, aerospace, biotech, and quantum computing while reducing reliance on foreign tech. It also boosts funding, with science spending set to rise by ~10% annually and overall R&D by at least 7%.

The UK government has announced up to £2 billion for quantum technologies, including more than £1 billion over the next four years, alongside a new procurement programme called ProQure to help scale quantum computing in the UK. The funding will support several areas: over £500 million for quantum computing, £125 million for quantum networking, and £205 million for quantum sensing and navigation, plus smaller allocations for research hubs, infrastructure, skills, and commercialisation.

The EU has opened a €180 million funding call to strengthen the resilience of subsea internet cables by supporting backup systems, alternative routes, and redundancy measures. The funding is meant to reduce the risk of outages and external threats to critical undersea infrastructure, reflecting the EU’s growing concern with digital resilience, cybersecurity, and technological sovereignty.

China has approved NEO, a brain–computer interface developed by Neuracle, for use beyond clinical trials to help people with severe paralysis regain hand movement. The implant reads brain signals when users imagine moving their hand and translates them into commands for a robotic glove, with early trial results showing improved ability to perform everyday tasks such as grasping, eating, and drinking.

Security

US President Donald Trump released his administration’s national cybersecurity strategy, outlining priorities across six policy areas: offensive and defensive cyber operations, federal network security, critical infrastructure protection, regulatory reform, emerging technology leadership (including in AI), and workforce development.

Trump also signed an executive order the same day, directing the attorney general to prioritise cybercrime prosecution, tasking agencies with reviewing tools to counter international criminal organisations, and assigning the Department of Homeland Security expanded training responsibilities. The strategy document spans five pages of substantive text, with administration officials describing it as intentionally high-level. The White House stated that more detailed implementation guidance would follow.

Pro-Iranian hacker group Handala claimed responsibility for a cyberattack on US medical device giant Stryker. The group has stated that the cyberattack is retaliation for a missile strike on an elementary school in Iran. Stryker confirmed the cyberattack in a statement, noting that order processing, manufacturing and shipping are disrupted, but that connected products have not been impacted. The FBI then seized four websites tied to Handala, the pro-Iranian hacking group that claimed responsibility for the attack, and to Iran’s Ministry of Intelligence and Security (MOIS).

Iran’s Revolutionary Guard has threatened to target major US tech companies, including Apple, Google, Meta, Microsoft, Intel, Oracle, Nvidia, Tesla, and Palantir, if more Iranian leaders are killed, accusing the companies of helping identify assassination targets. Iran’s army also claimed to have targeted Israeli communications, telecommunications and industrial centres in response to attacks on Iranian infrastructure.

Long constrained by a defensive security doctrine, Japan will introduce ‘hack-back’ powers from October. The change comes around as part of Japan’s ‘Active Cyber Defence’ law, which was passed in 2025 and is rolling out in incremental stages through 2027.

The EU has imposed sanctions over cyber attacks targeting its member states and partners, listing China-based Integrity Technology Group and Anxun Information Technology, as well as Iran-based Emennet Pasargad, along with Anxun’s co-founders. The sanctions entail an asset freeze and a travel ban for the listed individuals. The EU citizens and entities are additionally prohibited from making funds available to the designated companies.

Authorities in the Netherlands reported that hackers—believed to be linked to Russia—have launched large-scale phishing operations aimed at diplomats, military personnel, government officials, and journalists. Instead of breaking the apps’ encryption, attackers trick users into sharing verification codes or linking devices, allowing them to take over accounts and access sensitive conversations.

Portugal’s intelligence service has issued a similar alert, describing a global campaign by foreign state-backed actors seeking access to the messaging accounts of officials and others with privileged information. Once inside an account, attackers can read chats, access shared files, and use the compromised profile to target additional victims through further phishing attempts.

The EU has launched its ProtectEU counterterrorism agenda to strengthen preparedness against evolving threats, with a strong focus on how terrorists use digital tools such as social media, AI, encrypted platforms, crypto-assets, and drones. The plan combines stronger intelligence and Europol support, tougher enforcement of online content under the DSA, protection of public spaces and critical infrastructure, and closer international cooperation.

INTERPOL has launched a new global task force at the Global Fraud Summit 2026 as part of a more coordinated, data-driven response to the rapid global expansion of financial fraud. The task force is jointly developed by the UK’s Home Office and INTERPOL and is codenamed Operation Shadow Storm. The task force will target scam centres and their links to cybercrime and human trafficking, using tools such as stop-payment mechanisms and international intelligence-sharing networks. The initial focus of the task force will be dismantling criminal operations across Southeast Asia.

Simultaneously, major technology and consumer-facing companies, including Google, Amazon, Meta, and OpenAI, have signed the ‘Industry Accord Against Online Scams and Fraud’ at the Global Fraud Summit 2026. The companies pledged to focus on deploying proactive security measures and AI-driven detection systems; strengthening information sharing between industry and law enforcement to better identify and respond to fraud; enhancing resilience through advanced defensive technologies and rapid response mechanisms; and improving public education to help individuals recognise and avoid scams.

The EU has been unable to reach an agreement on extending temporary rules that allow online platforms to detect child sexual abuse material, leaving the current framework set to expire in April. The existing rules, in place since 2021, permit technology companies to voluntarily scan their services for harmful content, supporting efforts to identify and remove illegal material. But negotiations between the European Parliament and member states stalled over key issues — especially whether such measures should apply to encrypted services. Attention now shifts to the long-delayed permanent framework (the Child Sexual Abuse Regulation).

Brazil has started enforcing a new law aimed at strengthening protections for children online, marking a significant shift in how digital platforms are regulated in the country. The legislation, known as ECA Digital, introduces obligations such as age verification, stricter content moderation, and mechanisms to remove harmful material involving minors without requiring a court order. The law also targets platform design, requiring companies to limit features that may encourage compulsive use among children, such as excessive notifications, profiling for targeted advertising, and design elements that prolong user engagement. The law allows authorities to impose warnings and fines of up to $10 million for violations. In severe cases, courts may order the suspension or banning of platforms operating in Brazil.

Indonesia’s Communication and Digital Affairs Minister signed a government regulation that means children under 16 can no longer have accounts on high-risk digital platforms. This will reportedly include YouTube, TikTok, Facebook, Instagram, Threads, X, Bigo Live and Roblox. Implementation of the regulation will begin gradually from 28 March.

In Ecuador, the issue is framed in terms of security. A proposed ban on under-15s is linked to concerns that platforms are being used by criminal groups to contact and recruit minors. This shifts the rationale away from well-being and toward crime prevention, positioning social media restrictions as part of a broader security response.

A proposed social media ban for under-16s has been rejected by UK MPs, with 307 voting against and 173 in favour. However, a government-backed pilot is trialling different forms of restriction—full bans, time limits, and curfews— for six weeks. Participants will be interviewed before and after the trial to assess behavioural and practical outcomes, including how easily restrictions can be enforced and whether teenagers attempt to bypass controls.

Australia’s eSafety Commissioner states that platforms have removed or restricted millions of under-16 accounts under the country’s social media age ban, but serious compliance problems remain, including weak age-assurance systems and reporting tools that are hard for parents to use. Investigations into five major platforms are continuing, with enforcement decisions expected by mid-2026.

Austria plans to ban social media use for children under 14, joining a broader international move toward stricter youth online-safety rules. The government says the measure is meant to protect children from addictive platform design, violence, misinformation, and harmful beauty standards, and it also plans to add a new school subject on media and democracy to strengthen digital literacy.

France is considering a new law to ban social media for children under 15, while also proposing a digital curfew for older teens and extending school phone restrictions to high schools. The regulation reflects a broader push for stronger regulation of online harms affecting young people, including cyberbullying, harmful content, and excessive screen exposure, and aligns with similar child-safety measures already seen in countries such as Australia.

A Swiss survey found strong public mistrust of major tech companies such as Google, TikTok, and Meta, with most respondents viewing them as profit-driven, politically influential, and a source of dependence on foreign powers. At the same time, a majority still sees digitalisation as broadly positive, but wants the state to play a stronger role in ensuring that AI, algorithms, and digital platforms do not harm democracy or society.

Australia has begun enforcing new online child-safety rules that require platforms, including social media, app stores, gaming services, search engines, pornography sites, and AI chatbots, to use age-assurance measures and block minors from harmful or explicit content, including sexual and self-harm-related chatbot interactions. The eSafety Commissioner oversees the rules, and companies can face penalties of up to AUD 49.5 million per breach for non-compliance.

European Commission President Ursula von der Leyen convened the first meeting of the Special Panel on child safety online, announced in her 2025 State of the Union address. The panel will provide expert guidance on protecting and empowering children online and explore potential harmonised age limits for social media access. The panel aims to

Economic

The EU and Canada have begun negotiations on a Digital Trade Agreement to expand the digital side of their existing trade relationship, aiming to set clearer rules for cross-border digital commerce. The talks cover issues such as paperless trade, recognition of e-signatures and digital contracts, no customs duties on electronic transmissions, and limits on data-localisation and forced source-code transfer requirements, while still preserving governments’ ability to regulate the digital economy.

The EU and Australia have deepened ties through a new Security and Defence Partnership, the conclusion of free trade agreement negotiations, and the launch of talks on Australia’s accession to Horizon Europe. Together, these moves are meant to expand cooperation on cybersecurity, crisis response, AI and other emerging technologies, data flows, critical raw materials, and trade, signalling a broader strategic alignment beyond economics alone.

Australia is moving toward a national licensing regime for crypto exchanges and tokenisation platforms under its financial services framework, following a Senate committee’s recommendation to pass the Digital Assets Framework Bill 2025. The proposal would bring more of the crypto sector under formal regulation, though industry groups warn that broad definitions could unintentionally capture some infrastructure providers and wallet-related services.

Meta has announced that third-party AI chatbots will once again be allowed to operate through WhatsApp in Europe for a fee, reversing earlier restrictions that limited access to rival chatbot services on the platform. Under the new arrangement, companies will be able to distribute general-purpose AI chatbots via the WhatsApp Business API for 12 months. The change is intended to give European regulators time to complete their investigation while allowing competing AI services to operate within the platform ecosystem.

Google will overhaul its Play Store policies after settling a long-running dispute with Epic Games, creator of Fortnite. The changes include lowering in-app purchase commissions to 20%, adding a 5% fee for developers using Google’s billing system, and reducing subscription fees to 10%, alongside making it easier to install alternative app stores on Android. As part of the deal, Epic will return Fortnite to the Play Store while continuing to develop its own Android app store.

The WTO meeting ended without agreement on extending the e-commerce duty moratorium, while a group of members advanced a separate digital trade arrangement. Read more in our dedicated text.

The ECB has launched Appia, a roadmap for developing Europe’s tokenised financial markets, with Pontes as a DLT-based settlement solution linking tokenised-market infrastructure to the Eurosystem and enabling pilots from Q3 2026. The plan is meant to support the shift from traditional finance to tokenised markets while preserving financial stability, central bank settlement, and interoperability, and it is now open for public consultation.

A joint ILO–World Bank study finds that AI will affect jobs unevenly across 135 economies. Advanced economies face higher exposure, especially in clerical and professional work, while developing countries risk disruption without comparable productivity gains because they often lack the infrastructure, internet access, and skills needed to benefit. The report argues that outcomes will depend less on AI alone than on connectivity, training, job design, and social protections.

Legal

The World Data Organisation (WDO) was launched in Beijing as a new international non-profit platform focused on global data development and governance, with the stated aim of narrowing the global data divide, supporting the digital economy, and improving international cooperation on issues such as cross-border data flows, privacy, and security. The initiative reflects a broader push to make data governance a more structured part of global digital policymaking.

A Luxembourg court has annulled Amazon’s €746 million GDPR fine, not because the alleged privacy violations disappeared, but because it found the regulator’s penalty process was flawed, especially in how Amazon’s level of fault was assessed. The case will now return to Luxembourg’s data protection authority for reassessment.

Italy’s data protection authority has fined Intesa Sanpaolo €31.8 million after an employee repeatedly accessed thousands of customer accounts without authorisation, and the bank failed to detect it in time. Regulators said the case exposed serious weaknesses in internal monitoring, risk controls, confidentiality safeguards, accountability, and breach notification.

Development

Malta has launched the SMART Food project, a Malta–Italy initiative using AI and blockchain to build a digital platform that tracks food products from production to consumption. The aim is to improve traceability, transparency, safety, sustainability, and trust in the agri-food sector, while helping consumers and producers access real-time product information.

China has revised its rules for the 2026 national agricultural census, expanding the census to cover not only agriculture but also rural industrial development and village construction, while introducing new data-collection methods such as remote sensing. The updated rules also tighten data-quality controls, confidentiality obligations, and penalties for falsifying statistics, reflecting a stronger emphasis on both broader rural data collection and stricter state oversight.

The UK and the Philippines have agreed on a new partnership to expand digital education and edtech cooperation, combining UK expertise and investment support with Philippine education priorities. The initiative focuses on improving access to digital learning tools, skills development, and education technology, while strengthening broader bilateral ties in innovation and capacity-building.

Sociocultural

The first transparency reports under the EU’s Digital Services Act-linked Code of Conduct on Disinformation have been published, with signatories including major platforms and civil-society actors outlining measures they say they are taking against disinformation, especially around the war in Ukraine and election integrity. Their significance is that these are the first reports since the Code gained formal recognition under the DSA in February 2025, marking a shift from a mainly voluntary scheme to a more structured co-regulatory system based on commitments, reporting, and auditing.

The EU is reviewing X’s proposal to change its blue-check verification system after finding that paid verification without meaningful identity checks could mislead users under the Digital Services Act. X had been fined €120 million in December and given 60 working days to submit corrective measures, which the Commission is now assessing while the company also challenges the decision in court.

UNESCO has launched a South Africa-focused research initiative on the governance of harmful online content under its Social Media 4 Peace programme, supported by the EU, to study hate speech, disinformation, regulatory gaps, and platform governance. The aim is to produce practical, rights-based recommendations that strengthen digital governance, platform accountability, freedom of expression, and access to information in the local context.

Spain has launched HODIO, a digital tool to measure hate speech across social media. Combining AI, data analysis, and expert review, it will publish biannual reports ranking platforms by users’ exposure to harmful content, aiming to inform policymaking and pressure companies to act. However, critics have raised concerns about the transparency of HODIO and how authorities will define and classify hate speech, warning that poorly defined criteria could infringe on freedom of expression.

AI governance in March

National frameworks, strategies and guidelines

USA. The US government has unveiled a National AI Policy Framework outlining a comprehensive strategy for AI across federal agencies. The policy sets priorities for responsible AI development, data governance, workforce training and international collaboration, while emphasising ethical safeguards, public‑interest outcomes and national security. The framework also calls for accelerated investment in AI research and deployment, alongside coordinated oversight mechanisms to ensure transparency and accountability in federal AI systems.

Egypt. On 14 March 2026, Egypt published the National Guidelines for Trustworthy and Responsible AI. The Guidelines provide a national reference for the responsible development, deployment, and oversight of AI across public and private sectors, ensuring AI use is safe, ethical, and transparent while supporting innovation aligned with Egypt’s Vision 2030 and the National AI Strategy. Complementing the National AI Governance Framework, which defines what should be governed, these Guidelines specify how to comply, offering methodologies, metrics, and checklists to operationalise ethical principles. Targeted at data scientists, compliance officers, and developers, they provide actionable directions to protect individual rights, promote societal well-being, enhance accountability and transparency, and foster innovation grounded in safety. The guidelines also align Egypt with international standards and engage government entities, private enterprises, and community actors in responsible AI governance.

South Korea. South Korea has unveiled a national strategy to become one of the world’s top three AI powers by 2028. The plan combines investment in digital infrastructure, data systems and next-generation connectivity. Authorities aim to expand networks by advancing 5G capabilities and preparing for the commercial deployment of 6G by 2030. Cybersecurity and data integration are also key priorities to support a stronger digital ecosystem. The strategy includes developing talent across education levels and investing in core technologies such as semiconductors and quantum computing. AI adoption is expected to expand across sectors, including manufacturing, healthcare and agriculture.

Sovereignty

The EU. Tensions are emerging in the EU over AI infrastructure investment, with France, Poland, Austria, and Lithuania pushing to reserve part of the €20 billion AI Gigafactory project for European technologies, while Germany is sceptical about linking the project to digital sovereignty goals. Meanwhile, Germany is pursuing a major expansion of domestic data centres and AI processing power, supported by regulatory reforms, tax incentives, and land allocation to attract investment, aiming to reduce reliance on foreign providers.

Russia. The Russian government is proposing rules that could ban or restrict foreign AI tools such as ChatGPT, Claude and Gemini if they fail to store Russian user data domestically and comply with Moscow’s regulatory requirements. The proposals, from the Ministry for Digital Development, aim to extend Russia’s push for a sovereign internet, protecting citizens from ‘covert manipulation’ and enforcing ‘traditional Russian spiritual and moral values.’ Under the draft rules, cross-border AI systems that transmit user data abroad would face restrictions, whereas foreign models that can operate entirely within Russian infrastructure, such as Qwen or DeepSeek, could be deployed safely.

Content policy

The EU. The European Commission has released a second draft of its Code of Practice on marking and labelling AI-generated content, part of efforts to help companies comply with transparency requirements under Article 50 of the EU Artificial Intelligence Act. Section 1 of the code focuses on providers of generative AI systems and proposes a multi-layered approach to marking AI-generated content, including digitally signed metadata, imperceptible watermarking, and optional fingerprinting or logging. Providers are also expected to make detection tools available so users and authorities can verify whether content was generated or manipulated by AI. Section 2 addresses deployers of AI systems, requiring clear disclosure when deepfakes or AI-generated text intended to inform the public have been artificially generated or manipulated, using visible and accessible labels.

The European Council has endorsed proposals to ban AI from generating non-consensual sexual content (CSAM), adjust high-risk AI compliance timelines, and streamline the AI Act, including exemptions for some SMEs, registration requirements, and clarified oversight responsibilities. These moves reflect Europe’s broader effort to secure sovereign AI infrastructure and ensure safe, accountable AI deployment.

Netherlands, France. A Dutch court has ordered xAI and its Grok chatbot not to create or distribute non‑consensual sexual images. The judgement requires Grok’s operators to implement technical measures to block prompts or outputs capable of producing non‑consensual intimate imagery. The decision was framed as a necessary enforcement of personal rights and dignity in the digital age, setting a potentially influential precedent for European courts grappling with AI‑generated harm.

Meanwhile, the Paris prosecutor’s office said that the controversy surrounding sexually explicit deepfakes generated by Grok may have been deliberately amplified. The alleged reason was to artificially boost the value of X and xAI ahead of June 2026, when the new entity created by the merger between SpaceX and xAI is planned to be listed on the stock market.

Security

Australia. The eSafety Commissioner found that AI companion chatbots, including Character.AI, Nomi, Chai and Chub AI, are failing to protect children from harmful content, with weak safeguards against sexually explicit material and child sexual exploitation. Most platforms relied on self-declared age verification, lacked meaningful monitoring of AI inputs and outputs, and did not consistently provide links to crisis or mental health support. Commissioner Julie Inman Grant warned that as children increasingly use AI companions for emotional support, the absence of robust safety measures on self-harm, suicide and unlawful content poses serious risks, with non-compliance subject to civil penalties under Australia’s Age-Restricted Material Codes.

The UK. Secretary of State for Science, Innovation and Technology has called on online service providers to strengthen measures against digital harms targeting women and girls, as part of a commitment to halve such violence within a decade. The secretary called on tech companies to implement Ofcom’s guidance ‘A Safer Life Online for Women and Girls’, which outlines steps such as conducting risk assessments focused on women and girls, pre-launch abusability evaluations of features, strong default privacy settings, demonetising content promoting abuse, limiting the visibility of misogynistic content in search and recommendation feeds, and implementing rate limits to curb coordinated harassment. The guidelines should be implemented by the end of 2026 at the latest.

The USA. The US government is facing two lawsuits from AI firm Anthropic after the Pentagon designated the company a supply-chain risk, effectively barring its technology from defence contracts.

The Department of Justice argues the designation is lawful and grounded in national security, citing Anthropic’s refusal to allow its AI to be used for autonomous weapons and domestic surveillance. Anthropic, in turn, claims the move is unlawful and retaliatory, targeting its policy positions rather than any genuine security risk.

In the California case, a federal judge has temporarily blocked the government from enforcing the designation. The court found that Anthropic’s conduct does not meet the legal threshold under Section 3252, which is limited to covert adversarial threats such as sabotage or system subversion—not public stances or contract disputes. The ruling also highlights procedural failures, including insufficient risk assessment, lack of interagency consultation, and failure to consider less restrictive measures.

The judge further raised constitutional concerns, noting the designation may have been influenced by Anthropic’s speech and that the company was likely denied due process. Evidence of immediate and significant harm—lost contracts, reputational damage, and disrupted business relationships—justified granting a preliminary injunction, though a final ruling may take months.

In parallel, Anthropic is pursuing a second case in Washington, D.C., challenging its supply chain designation before a three-judge panel at the D.C. Circuit Court of Appeals, specifically contesting the legal authority invoked under the Federal Acquisition Supply Chain Security Act (FASCA).The legal dispute has drawn support from across the tech sector, with companies including Microsoft, Google, Amazon and OpenAI backing Anthropic’s legal challenge through amicus filings. Industry leaders warn that the government’s designation could set a precedent that destabilises the US AI ecosystem and disrupts suppliers working with both government and private-sector AI systems.

Pay to think: Intelligence on a meter

‘We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter and use it for whatever they want to use it for,’ Sam Altman, CEO of OpenAI, recently stated.

On the surface, this could sound like a vision of empowerment: on-demand access to superhuman reasoning, available to anyone with enough money to buy it. But Altman’s metaphor is precise. Utilities are not owned by the public; they are controlled by powerful providers who set the rates, terms, and infrastructure.

Our knowledge is already becoming commodified by tech companies and the advertising industry. But what OpenAI’s CEO suggests is a world in which intelligence itself is outsourced to a handful of platforms.

The AI monopolisation of intelligence challenges one of the pillars of civilisation built over millennia: That knowledge defines what it means to be human.

Altman is therefore not just describing a business model; he is also outlining a new social order, one in which intelligence is centralised, privatised, and sold back to humanity by major AI companies.

Person, Face, Head, Clothing, Coat, Chanakya

Not an inevitable future. The battle for human intelligence and knowledge – for who owns the capacity to think, to know, to decide – is not yet over.

The real alternative to monopolising and metering our knowledge back to us isn’t no AI; the real alternative is to have AI as an extension of our personal knowledge shared communities, countries, and humanity, as per our preferences.

Communities, universities, companies, and countries can build bottom-up AI rooted in their own languages, values, and knowledge systems. Open-source models have made human-centred AI technically possible and financially affordable. This would lead to a distributed ecosystem in which AI strengthens human communities rather than subordinates them.

This text is an adaptation of Dr Jovan Kurbalija’s blogpost ‘The war we’re not watching: The fight for the future of human knowledge.’

From content to design: Juries signal new era of accountability for tech giants

Two recent US jury verdicts are beginning to redraw the boundaries of responsibility for social media platforms, with implications that extend well beyond the individual cases.

In New Mexico, a jury ordered Meta to pay $375 million after finding it misled users about the safety of its platforms for children. The lawsuit, brought by Attorney General Raul Torrez, accused Meta of violating the state’s consumer protection laws by misrepresenting how safe its platforms are for minors while building features and algorithms that, in prosecutors’ view, entice prolonged use and expose children to significant risks. Those risks include addiction-like engagement, exposure to harmful sexual content, unwanted private communications with adults, sleep disruption from compulsive use, and environments where predators can operate with relative ease. Jurors were presented with internal research and testimony from former employees, including whistle-blower Arturo Béjar, suggesting the company was aware of these risks but failed to adequately warn the public or mitigate harm. Meta has rejected the verdict and plans to appeal.

Simultaneously, a Los Angeles jury reached a related conclusion in a different context. It found Meta and YouTube—owned by Google—negligent in the design and operation of their platforms in a case focused on social media addiction. The lawsuit, brought by a young woman identified as K.G.M., argued that compulsive use of these platforms during her teenage years contributed to depression, anxiety, and body dysmorphia. The jury agreed, awarding $6 million in damages and assigning 70% of the liability to Meta and 30% to Google. Both companies have said they will appeal, maintaining that mental health outcomes cannot be attributed to a single platform.

Why does it matter? The financial penalties in these cases are small for companies of this scale. The broader significance of the verdicts lies elsewhere.

Adult, Female, Person, Woman, Face, Head, Publication, Book

Historically, platforms have relied on legal protections—most notably Section 230 of the US Communications Act—to shield themselves from liability for user-generated content. These rulings, however, begin to test a different theory: that liability can arise not just from what users post, but from how platforms structure, recommend, and amplify content.

This distinction matters because it targets the core of the modern social media business model. Platforms like Meta and Google are built around maximising user engagement—time spent, interactions, and content consumption—which in turn drives advertising revenue. To achieve this, they rely on recommendation systems, frictionless interfaces, and behavioural design features such as autoplay, infinite scroll, and push notifications. These are not incidental elements; they are foundational to how platforms retain users and monetise attention.

The emerging legal argument is that some of these design choices may actively contribute to harm, particularly for minors. In the New Mexico case, the focus was on exposure to harmful and exploitative content. In Los Angeles, the emphasis was to compulsive use and its mental health effects. But both cases converge on a similar point: that platform architecture itself—not just isolated content failures—can create foreseeable risks.

If this reasoning gains traction in courts, it introduces a new kind of pressure on technology companies. The issue is not the size of any single fine, but the cumulative effect of thousands of similar lawsuits, rising compliance costs, and the possibility of precedent-setting rulings that reshape acceptable design practices. Engagement-maximising systems, long treated as a competitive advantage, could become a source of legal vulnerability.

That creates a structural tension. Reducing harmful outcomes may require dialling back precisely those features that make platforms so effective at capturing attention. Even modest declines, when applied across billions of users, can translate into significant revenue impacts.

The path forward. Companies are unlikely to abandon their core models outright. A more probable response is adaptation. This could include re-optimising algorithms toward safer forms of engagement, segmenting products by age with stricter defaults for minors, and investing in more robust safety and audit mechanisms. There may also be a gradual shift toward alternative revenue streams—such as subscriptions, creator monetisation, or commerce integrations—to reduce reliance on pure attention-based advertising.

Legal strategy will also play a role. Both Meta and Google are appealing these verdicts, and future rulings will determine how far courts are willing to go in attributing harm to design choices. Companies are likely to strengthen disclosures, expand parental controls, and document internal risk assessments to demonstrate due diligence. Such measures may not eliminate liability, but they can shape how responsibility is interpreted.

Ultimately, the key question is whether these cases represent isolated outcomes or the beginning of a broader legal shift.

UN kicks off Global Mechanism on ICT security, road ahead murky

After almost three decades of stop-start cybersecurity negotiations at the UN, the long-anticipated Global Mechanism on ICT security has finally kicked off.

It is the first permanent forum of its kind since discussions on ICT security began back in 1998, and its mere existence says a lot about how far those talks have come.

But if the launch felt like a breakthrough, the organisational session quickly brought things back down to earth. Beyond what was already sketched out in Annex C and the OEWG’s Final Report, it remained unclear how the Mechanism would actually organise itself in practice.

The session raised plenty of questions—about structure, priorities, and process—but offered few real answers, leaving the sense that while the Mechanism now exists, what it will do and how it will do it is still very much up for grabs.

A new body, a new mandate, and a newly elected Chair, Egriselda López of El Salvador, injected renewed optimism into the Global Mechanism’s first organisational session. Yet, within minutes, it became evident that the Global Mechanism did not start with a blank slate, but rather inherited the OEWG’s long list of disagreements.

Russia opened the discussion by disputing the legitimacy of the Chair nomination, which they claimed was guided solely by the UNODA and thus limited state participation in the process. They used this opportunity to stress that all decisions under the new process must be based on consensus and be completely intergovernmental.

The substantive issues on the agenda

For the provisional agenda of the mechanism’s July session, the Chair circulated a draft agenda organised around the five pillars of the framework for responsible state behaviour in the use of ICTs. However, Iran and Russia argued that the wording of agenda item 5 did not precisely reflect paragraph 9 of Annex C of the OEWG final report and called for correction at this session. The EU and Canada rejected this, arguing the draft already referenced all relevant documents and that isolating one paragraph would itself constitute renegotiation. The USA reserved its position entirely, preferring that the July plenary adopt its own agenda. No consensus was reached, and the Chair will continue consultations before July.

The mechanism inherited many unresolved substantive debates from its predecessors.

On international law, there is widespread agreement that considerable work remains to be done, but little agreement on how to carry it out. The majority of delegations have shown clear support for strengthening the existing normative framework and reaffirming the UN Charter’s application to cyberspace.

A broad majority of states expressed support for ensuring that the mechanism remains action-oriented, with a strong focus on practicality and the implementation of agreed frameworks on international law, norms, CBMs, and capacity-building (Chile, Nauru, Portugal, Switzerland, the United Kingdom, Estonia, Italy, Australia, the Democratic Republic of the Congo, Antigua and Barbuda, Sudan, Vanuatu, Albania, Vietnam, India, Greece, Rwanda, the Dominican Republic, North Macedonia, Kiribati).

In particular, some delegations advocated for applying the framework to concrete scenarios as a way to stimulate implementation (Japan, the Netherlands, the United Kingdom, Sudan). China was the only delegation to emphasise that further development of the framework is equally important alongside its implementation.

The EU highlighted the norm checklist, a hotly debated issue in the previous mechanism, as an area for further improvement.

However, to many states, a fundamental concern remains. Capacity building initiatives risk stalling without reliable funding, so many delegations, primarily from developing countries, urged the Global Mechanism to prioritise the operationalisation of the UN Voluntary Fund, which was tabled but left unresolved by the OEWG.

Dedicated thematic groups: Who, what and how

The often broad agenda and long-winded statements of delegations in OEWG plenary sessions left little room for technical depth, leaving many delegations frustrated with the gap between consensus language and concrete action.

The Dedicated thematic groups (DGTs) were created to address this issue precisely by setting up an informal, technical forum to advance practical initiatives already agreed on, such as the Global ICT Security Cooperation and Capacity Building Portal. However, the practicalities on how they should be set up and administered are going to be hotly contested as it will influence what gets on the agenda, who drives it, and whether this new system is capable of delivering real outcomes over time.

Who will lead DTGs?

The dominant and most contested question of the session was who would appoint the co-facilitators for the two Dedicated Thematic Groups. The Chair proposed appointing two co-facilitators per DTG: one from a developed country, one from a developing country, drawing on GA practice, under which the Chair appoints co-facilitators for intergovernmental processes. She indicated her intention to hold broad informal consultations before making appointments, and committed to geographic balance, gender parity where practicable, and relevant technical expertise as selection criteria.

Who ends up in these roles matters considerably: the co-facilitators will steer the DTG discussions, shape their agendas, and channel recommendations to the plenary.

A broad coalition of states supported the Chair’s approach, including the EU, speaking on behalf of its member states and several aligned countries such as France, Germany, Australia, the United Kingdom, the Netherlands, Switzerland, Japan, Egypt, Senegal, Nigeria, Malaysia, Moldova, and others. Egypt and Senegal were among the most direct, noting that delays in operationalising the mechanism would waste the intersessional period and erode its credibility, particularly for developing countries eager to move from procedure to substance.

Another group of states, led by Russia and supported by Iran, China, Belarus, Nicaragua, and Cuba, argued that co-facilitator appointments must be approved by member states by consensus rather than made unilaterally by the Chair. Russia contended that DTG co-facilitators handle substantive political matters and therefore constitute officials whose appointment requires a collective agreement. Russia also raised a geographic argument: assigning one developed-country and one developing-country co-facilitator per DTG still disproportionately favours developed states, which represent less than one-fifth of UN membership. Iran added that the early OEWG draft text had explicitly authorised the Chair to appoint DTG facilitators, but that this provision was deliberately removed during negotiations, signalling a lack of agreement on the matter.

The Chair affirmed her intention to consult all member states informally before presenting candidates and called on delegations to show flexibility given the urgency of getting the mechanism’s work underway. Russia subsequently stated its understanding that candidates would be determined through broad consultation, followed by consensus-based approval, but the Chair neither confirmed nor rejected this interpretation.

The question is effectively deferred to the intersessional period, meaning the composition of the DTG leadership teams remains unresolved and will require continued diplomatic engagement before July.

What will DTGs discuss?

A closely related debate concerned who decides what the DTGs will actually discuss. Several Western and like-minded delegations (e.g., Germany, France, Canada, the United Kingdom, and Australia) highlighted that it is a prerogative of the Chair and co-facilitators, to be exercised in close consultation with states. These delegations proposed ransomware and critical infrastructure protection as natural starting points, citing their frequency across national statements and OEWG discussions.

Iran and Russia emphasised that topics must be determined by consensus among all member states. Argentina argued that the plenary should maintain control over the agenda rather than ceding too much responsibility to the co-facilitators.

Morocco instead advocated a bottom-up model in which DTGs define their own priority subtopics from the start, based on member states’ expressed preferences to maintain regional balance and ownership.

In this sense, the DTGs’ credibility hinges on a delicate balance, having to be ambitious enough to move conversations into action but also focused enough on issues with broad support so that their outputs survive in plenary.

No decision was taken. For industry and civil society organisations with specific thematic priorities, this remains an active opening: states are currently receptive to input on which topics the DTGs should prioritise.

Colombia put forward a process proposal that drew broadly positive reactions across delegations. It recommended that:

DTG mandates be time-limited with clearly defined and measurable outputs;
DTG 1 addresses specific rotating subjects rather than its entire mandate simultaneously, and
DTG outputs systematically distinguish between recommendations on which consensus exists and those still under development.

Senegal made a complementary point: reports should document both areas of agreement and divergence, preserving a record of discussions even when no consensus was reached. Both proposals reflect a wider concern that, without structured outputs and clear timelines, the mechanism risks reproducing the open-ended deliberation of the OEWG without generating implementable results.

How will DTGs feed into the plenary?

Another issue discussed was how DGT work feeds into plenary work. Brazil made it clear that without a defined protocol for elevating DTG reports to the plenary and formally accepting their recommendations, the groups risk becoming talking shops that are disconnected from the mechanism’s official conclusions. Their proposed solution, which still has to achieve support, is to keep DGT conversations primarily informal but include a short formal section for decision-making.

Stakeholder participation

A long-standing point of contention and possibly the most politically-charged was the role of non-governmental actors in the groups. The effective participation of interested stakeholders remains uncertain.

Some delegations adopted a more accommodating stance, recognising that stakeholders can enhance the quality of deliberations (Sudan, Antigua and Barbuda) and contribute to more practical outcomes (Vietnam, Dominican Republic), while underscoring the importance of preserving the intergovernmental nature of the process (Sudan, Vietnam).

Canada and like-minded states argued that the July 2025 consensus clearly provides for states to nominate experts for DTG briefings and for the wider stakeholder community to participate throughout DTG discussions.

Iran contested this, asserting that stakeholder modalities agreed for the mechanism apply equally to DTGs. Russia also argued that expert briefings from external stakeholders are a possibility rather than a standard feature, and that inviting external briefers requires member-state agreement on a case-by-case basis.

How this is resolved will directly determine the degree of access the private sector, technical community, and civil society organisations have to the DTG process in practice.

What’s next?

The session closed without resolution on its two most consequential questions: co-facilitator appointments and the provisional plenary agenda. The Chair will convene informal intersessional consultations on both and issue a programme of work document before July in all UN languages.

The Secretariat will open an annual stakeholder accreditation window in the coming weeks; stakeholders wishing to participate in plenary sessions and review conferences can monitor the Digital Watch Observatory web page, where we track the process, for details.

The broader tension remains unresolved, and how it is managed in the intersessional period will largely determine whether the July plenary can open with the mechanism’s operational foundations in place.

The Chair also confirmed the two key dates for 2026:

the substantive plenary session scheduled for 20-24 July, and
the DTG meetings scheduled for 7-11 December, both at UN Headquarters in New York

For stakeholders tracking or seeking to contribute to these discussions, these are the dates to plan around.

Deadlock at WTO: Moratorium lapse meets plurilateral momentum

At the 14th Ministerial Conference of the World Trade Organization (MC14) in Yaoundé, Cameroon, digital trade dominated the agenda through two parallel tracks—each pointing in a different direction and illustrating both the limits and evolution of the multilateral system.

The moratorium on customs duties on electronic transmissions. The long-standing moratorium—renewed every two years since 1998—expired on 31 March after members failed to reach consensus on the length of a new extension, with differing views among members preventing a deal.

While some members, particularly the USA, sought a longer-term solution, others have traditionally advocated a shorter renewal period, reflecting a desire for caution given the rapid pace of technological change and the need to preserve policy flexibility for the future.

During MC14, Brazil was the leading voice, emphasising the importance of caution in light of developments such as AI and 3D printing, suggesting that a shorter extension with room for review would allow members to reassess as the digital landscape evolves. Efforts to find a middle ground ultimately fell short as time ran out.

The outcome also meant that a broader set of discussions on WTO reform, which had been politically linked to the approval of the moratorium, remained unresolved.

This is not the first time the moratorium lapsed; it happened at the 1999 Seattle ministerial, before the moratorium was reinstated at Doha two years later. The current expiry of the moratorium does not mean tariffs will automatically be imposed.

Still, it creates policy space for some countries to consider introducing tariffs if they are not bound by trade agreements that prohibit customs duties on electronic transmissions.

Plurilateral Agreement on E-commerce. In parallel, however, a different dynamic unfolded. A coalition of 66 WTO members announced they would move forward with implementing the plurilateral Agreement on Electronic Commerce concluded in 2024 by the Joint Statement Initiative on e-commerce (JSI), through interim arrangements.

Reminder: WTO Joint Statement Initiatives (JSIs) are a way for a group of World Trade Organization members to move forward on specific issues without waiting for the entire organisation to reach a consensus. They are open to any WTO Member.

Australia, Japan, and Singapore, serving as co-convenors of the JSI on e-commerce, confirmed that the pact, which aims to facilitate digital trade and prohibit duties on e-commerce transactions, will enter into force once 45 members have formally notified their acceptance.

People, Person, Adult, Male, Man, Crowd, Face, Head, Baby, Book, Publication, Jury

What’s next for e-commerce discussions? Discussions on the moratorium, the WTO reform, and the future of the Work Programme on e-commerce (WPEC) are expected to continue at the next General Council meeting in May in Geneva.

In the meantime, JSI members will continue to seek inclusion of the Agreement under the WTO legal architecture.

The JSIs and their outcomes face opposition from a number of WTO members. The JSI themselves, these countries argue, lack legal status because they were not launched by consensus. Similarly, these countries claim that the outcomes of JIs are not based on consensus and are neither multilateral agreements nor plurilateral agreements as defined in Article IV of the agreement that established the WTO – the Marrakesh Agreement.

For instance, India registered dissent against the incorporation of the agreement achieved within another plurilateral negotiation, on Investment Facilitation for Development, into the WTO rulebook.

The country argued that incorporating such frameworks into the WTO rulebook risks eroding the organisation’s foundational principles. It asked for a discussion of guardrails and legal safeguards before integrating any specific plurilateral outcome into the WTO.

Last month in Geneva

The Data Technology Seminar 2026, organised by the European Broadcasting Union, took place from 10 to 12 March in Geneva. The event brought together media professionals and technology experts to discuss how AI and data systems are being developed, governed, and deployed in public service media. Sessions will explore topics such as AI strategy and governance, metadata platforms, hybrid search, audience personalisation, and the use of generative AI in editorial and production workflows.

The World Intellectual Property Organization (WIPO) launched the AI Infrastructure Interchange (AIII) on 17 March in Geneva and online. The programme included keynote remarks, panel discussions, and presentations addressing the role of technical collaboration between creators, rightsholders, and technology companies. Participants also discussed the objectives of the AIII initiative and the establishment of a Technical Exchange Network intended to support ongoing expert dialogue on practical challenges and opportunities.

The Geneva Graduate Institute organised a briefing lunch on 23 March to examine evolving transatlantic dynamics at the intersection of US politics and the global influence of major technology platforms. The discussion explored how recent political developments in the USA and the concentration of technological power shape Europe’s position, including questions of dependency, regulation, and strategic autonomy.

The International Labour Organization (ILO) hosted a session on the macroeconomic impacts of AI on 25 March, showcasing a new World Bank Group model that treats AI as a structural transformation of production. The tool simulates how AI adoption affects sectors, occupations, and prices, helping policymakers assess implications for growth, equity, and structural change. A first case study in Poland will explore its application, with potential use in other emerging and middle-income economies.

On 30 and 31 March, the International Telecommunications Union (ITU) held a two-day workshop on ‘Trustable and Interoperable Digital Identities for Human and Agentic AI’ in Geneva. It btought together stakeholders from governments, industry, academia, and standards bodies to examine technical approaches related to trust frameworks, trust management, security, and interoperability; and to investigate actionable recommendations and consolidated insights to advance standardisation work in the field.

The Inter-Parliamentary Union (IPU) hosted a webinar on ‘Building AI Literacy in Parliaments‘ on Wednesday, 1 April 2026, to explore how parliaments can develop training and resources to support AI literacy among members, parliamentary staff, and IT teams. The webinar will highlight the IPU Guidelines for AI in parliaments, emphasising that AI literacy should reach all roles within parliaments.

Weekly #255 From content to design: Juries signal new era of accountability for tech giants

20 – 27 March 2026

HIGHLIGHT OF THE WEEK

From content to design: Juries signal new era of accountability for tech giants

Two recent US jury verdicts are beginning to redraw the boundaries of responsibility for social media platforms, with implications that extend well beyond the individual cases.

In New Mexico, a jury ordered Meta to pay $375 million after finding it misled users about the safety of its platforms for children. The case, brought by Attorney General Raul Torrez, centred on evidence that services such as Facebook and Instagram exposed minors to sexually explicit content and potential predators. Jurors were presented with internal research and testimony from former employees, including whistle-blower Arturo Béjar, suggesting the company was aware of these risks but failed to adequately warn the public or mitigate harm. Meta has rejected the verdict and plans to appeal.

Simultaneously, a Los Angeles jury reached a related conclusion in a different context. It found Meta and YouTube—owned by Google—negligent in the design and operation of their platforms in a case focused on social media addiction. The lawsuit, brought by a young woman identified as K.G.M., argued that compulsive use of these platforms during her teenage years contributed to depression, anxiety, and body dysmorphia. The jury agreed, awarding $6 million in damages and assigning 70% of the liability to Meta and 30% to Google. Both companies have said they will appeal, maintaining that mental health outcomes cannot be attributed to a single platform.

Why does it matter? The financial penalties in these cases are small for companies of this scale. The broader significance of the verdicts lies elsewhere.

Historically, platforms have relied on legal protections—most notably Section 230 of the US Communications Act—to shield themselves from liability for user-generated content. These rulings, however, begin to test a different theory: that liability can arise not just from what users post, but from how platforms structure, recommend, and amplify content.

This distinction matters because it targets the core of the modern social media business model. Platforms like Meta and Google are built around maximising user engagement—time spent, interactions, and content consumption—which in turn drives advertising revenue. To achieve this, they rely on recommendation systems, frictionless interfaces, and behavioural design features such as autoplay, infinite scroll, and push notifications. These are not incidental elements; they are foundational to how platforms retain users and monetise attention.

The emerging legal argument is that some of these design choices may actively contribute to harm, particularly for minors. In the New Mexico case, the focus was on exposure to harmful and exploitative content. In Los Angeles, the emphasis was to compulsive use and its mental health effects. But both cases converge on a similar point: that platform architecture itself—not just isolated content failures—can create foreseeable risks.

If this reasoning gains traction in courts, it introduces a new kind of pressure on technology companies. The issue is not the size of any single fine, but the cumulative effect of thousands of similar lawsuits, rising compliance costs, and the possibility of precedent-setting rulings that reshape acceptable design practices. Engagement-maximising systems, long treated as a competitive advantage, could become a source of legal vulnerability.

That creates a structural tension. Reducing harmful outcomes may require dialling back precisely those features that make platforms so effective at capturing attention. Even modest declines, when applied across billions of users, can translate into significant revenue impacts.

The path forward? Companies are unlikely to abandon their core models outright. A more probable response is adaptation. This could include re-optimising algorithms toward safer forms of engagement, segmenting products by age with stricter defaults for minors, and investing in more robust safety and audit mechanisms. There may also be a gradual shift toward alternative revenue streams—such as subscriptions, creator monetisation, or commerce integrations—to reduce reliance on pure attention-based advertising.

Legal strategy will also play a role. Both Meta and Google are appealing these verdicts, and future rulings will determine how far courts are willing to go in attributing harm to design choices. Companies are likely to strengthen disclosures, expand parental controls, and document internal risk assessments to demonstrate due diligence. Such measures may not eliminate liability, but they can shape how responsibility is interpreted.

Ultimately, the key question is whether these cases represent isolated outcomes or the beginning of a broader legal shift.

IN OTHER NEWS LAST WEEK

This week in AI governance

USA. The US government has unveiled a National AI Policy Framework outlining a comprehensive strategy for AI across federal agencies. The policy sets priorities for responsible AI development, data governance, workforce training and international collaboration, while emphasising ethical safeguards, public‑interest outcomes and national security. The framework also calls for accelerated investment in AI research and deployment, alongside coordinated oversight mechanisms to ensure transparency and accountability in federal AI systems.

Netherlands, France. A Dutch court has ordered xAI and its Grok chatbot not to create or distribute non‑consensual sexual images. The judgement requires Grok’s operators to implement technical measures to block prompts or outputs capable of producing non‑consensual intimate imagery. The decision was framed as a necessary enforcement of personal rights and dignity in the digital age, setting a potentially influential precedent for European courts grappling with AI‑generated harm.

Meanwhile, the Paris prosecutor’s office said that the controversy surrounding sexually explicit deepfakes generated by Grok may have been deliberately amplified. The alleged reason was to artificially boost the value of X and xAI ahead of June 2026, when the new entity created by the merger between SpaceX and xAI is planned to be listed on the stock market.

The UK. Secretary of State for Science, Innovation and Technology has called on online service providers to strengthen measures against digital harms targeting women and girls, as part of a commitment to halve such violence within a decade. The secretary called on tech companies to implement Ofcom’s guidance ‘A Safer Life Online for Women and Girls’, which outlines steps such as conducting risk assessments focused on women and girls, pre-launch abusability evaluations of features, strong default privacy settings, demonetising content promoting abuse, limiting the visibility of misogynistic content in search and recommendation feeds, and implementing rate limits to curb coordinated harassment. The guidelines should be implemented by the end of 2026 at the latest.

Australia. The eSafety Commissioner found that AI companion chatbots, including Character.AI, Nomi, Chai and Chub AI, are failing to protect children from harmful content, with weak safeguards against sexually explicit material and child sexual exploitation. Most platforms relied on self-declared age verification, lacked meaningful monitoring of AI inputs and outputs, and did not consistently provide links to crisis or mental health support. Commissioner Julie Inman Grant warned that as children increasingly use AI companions for emotional support, the absence of robust safety measures on self-harm, suicide and unlawful content poses serious risks, with non-compliance subject to civil penalties under Australia’s Age-Restricted Material Codes.

Russia. The Russian government is proposing rules that could ban or restrict foreign AI tools such as ChatGPT, Claude and Gemini if they fail to store Russian user data domestically and comply with Moscow’s regulatory requirements. The proposals, from the Ministry for Digital Development, aim to extend Russia’s push for a sovereign internet, protecting citizens from ‘covert manipulation’ and enforcing ‘traditional Russian spiritual and moral values.’ Under the draft rules, cross-border AI systems that transmit user data abroad would face restrictions, whereas foreign models that can operate entirely within Russian infrastructure, such as Qwen or DeepSeek, could be deployed safely.

Operationalising restrictions on children’s use of social media

Restrictions on children’s use of social media are rapidly moving from political debate into policy design.

In Ecuador, the issue is framed in terms of security. A proposed ban on under-15s is linked to concerns that platforms are being used by criminal groups to contact and recruit minors. This shifts the rationale away from well-being and toward crime prevention, positioning social media restrictions as part of a broader security response.

The UK is not yet legislating; it is testing. A government-backed pilot is trialling different forms of restriction—full bans, time limits, and curfews— for six weeks. Participants will be interviewed before and after the trial to assess behavioural and practical outcomes, including how easily restrictions can be enforced and whether teenagers attempt to bypass controls. The UK’s approach reflects a lack of evidence on effectiveness, despite growing political pressure to act

At the same time, UK regulators are addressing a core constraint: how to implement age checks without undermining privacy. The Information Commissioner’s Office and Ofcom have issued joint guidance clarifying how age assurance should comply with both the Online Safety Act and data protection law. The key signal is not new obligations, but integration—age verification systems must be designed to deliver safety and privacy simultaneously, rather than treating them as competing requirements.

This implementation challenge is where Brazil is also moving decisively. The National Data Protection Agency (ANPD) published preliminary guidelines for implementing the Digital ECA. The guidance requires platforms such as social media, gaming, and adult content services to move beyond self-declared age checks and implement more robust verification systems, with penalties for non-compliance reaching up to 50 million reais or 10% of local revenue. Final rules are expected in August 2026 following public consultation.

EU disinformation code signatories publish first reports under DSA

Signatories to the EU Code of Conduct on Disinformation have published new transparency reports describing the measures they say they are taking to reduce the spread of disinformation online.

Dedicated sections in the reports cover responses to ongoing crises, notably the conflict in Ukraine, as well as measures intended to safeguard the integrity of elections. Data on the implementation of disinformation-related measures is also included, alongside developments in signatories’ policies, tools, and partnerships under the Digital Services Act framework.

The reports are available through the Code’s Transparency Centre and come from a broad group of signatories, including online platforms such as Google, Meta, Microsoft, and TikTok, as well as fact-checkers, research organisations, civil society bodies, and representatives of the advertising industry.

Why does it matter? The reports are the first ones submitted since the Code was recognised as a code of conduct under the Digital Services Act in February 2025. A more formal role now applies to the Code than under its earlier voluntary setup: By placing the disinformation Code inside the Digital Services Act framework, the Commission and the Board are using voluntary commitments, transparency reporting, and auditing as part of a co-regulatory approach to systemic online risks.

Interim trade deal plausible at the WTO

Signatories to the E-Commerce Agreement, negotiated under the WTO Joint Statement Initiative (JSI), are planning to implement the deal on an interim basis despite continued opposition. At least 70 of the 72 countries that endorsed the agreement are expected to sign a declaration to that effect at the WTO Ministerial Conference (MC14) in Yaoundé, which kicked off yesterday.

The move comes as JSI members seek to advance the agreement despite the lack of consensus among the full WTO membership for its incorporation into the Organization’s Annex 4, a step that would require the support of all WTO members. The interim arrangement would take the form of a legally binding treaty among the signatories, expiring upon formal integration into the WTO framework.

The E-commerce Agreement, finalised in July 2024, includes provisions on trade facilitation (e-signatures, paperless trade, single window), personal data protection, and a commitment to refrain from imposing customs duties on electronic transmissions. The latter clause would ensure the continuation of duty-free e-commerce among signatories regardless of the outcome of the broader WTO moratorium on customs duties on electronic transmissions.

View all updates

LOOKING AHEAD

The organisational session of the Global Mechanism on developments in the field of ICTs in the context of international security and advancing responsible State behaviour in the use of ICTs will be held on 30-31 March (Monday and Tuesday) in New York, USA. This session marks the start of the substantive work of the Global Mechanism, a new single-track, permanent forum on ICT security under UN auspices. Diplo and the GIP will provide reporting and expert insights from the session—bookmark our dedicated page on the Digital Watch Observatory to stay informed.

Also on Monday and Tuesday, ITU will hold a two-day workshop on ‘Trustable and Interoperable Digital Identities for Human and Agentic AI’ in Geneva. It will bring together stakeholders from governments, industry, academia, and standards bodies to examine technical approaches related to trust frameworks, trust management, security, and interoperability; and to investigate actionable recommendations and consolidated insights to advance standardisation work in the field. The event is open to ITU member states, sector members, associates, academic institutions, and other interested participants at no cost, but registration is required.

The Inter-Parliamentary Union will host a webinar on ‘Building AI Literacy in Parliaments‘ on Wednesday, 1 April 2026, to explore how parliaments can develop training and resources to support AI literacy among members, parliamentary staff, and IT teams. The webinar will highlight the IPU Guidelines for AI in parliaments, emphasising that AI literacy should reach all roles within parliaments.

READING CORNER

BLOG featured image 2026 37 WTO Ministerial

Digital economy at a crossroads: WTO Ministerial and the future of digital trade – Diplo

Digital trade is growing faster than traditional trade, but governance is struggling to keep up. At the WTO, key decisions could redefine global economic power. Carolina von der Weid, James Görgen and Marilia Macilel debate what’s at stake and who decides.

The unexpected gift: What open-weight AI brings to smaller countries – Diplo

The AI race isn’t just for the tech giants. Open-weight AI gives smaller countries a valuable new opportunity. With access to flexible and powerful models, these nations can protect their digital independence and adapt technology for their own languages and cultures, argues Slobodan Kovrlija.

Weekly #254 Pay to think: Intelligence on a meter

13 March – 20 March 2026

HIGHLIGHT OF THE WEEK

Pay to think: Intelligence on a meter

This week’s highlight is, in fact, more of a lowlight. Last Friday, Sam Altman, CEO of OpenAI, remarked:

‘We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter and use it for whatever they want to use it for.’

Our knowledge is already becoming commodified by tech companies and the advertising industry. But what OpenAI’s CEO suggests is a world in which intelligence itself is outsourced to a handful of platforms.

The AI monopolisation of intelligence challenges one of the pillars of civilisation built over millennia: That knowledge defines what it means to be human.

Altman is therefore not just describing a business model; he is also outlining a new social order, one in which intelligence is centralised, privatised, and sold back to humanity by major AI companies.

Not an inevitable future. The battle for human intelligence and knowledge – for who owns the capacity to think, to know, to decide – is not yet over.

The real alternative to monopolising and metering our knowledge back to us isn’t no AI; the real alternative is to have AI as an extension of our personal knowledge shared communities, countries, and humanity, as per our preferences.

Communities, universities, companies, and countries can build bottom-up AI rooted in their own languages, values, and knowledge systems. Open-source models have made human-centred AI technically possible and financially affordable. This would lead to a distributed ecosystem in which AI strengthens human communities rather than subordinates them.

IN OTHER NEWS LAST WEEK

This week in AI governance

The USA. The Trump administration is defending the Pentagon’s decision to cut ties with Anthropic. The Department of Justice (DoJ) urged a federal judge to reject Anthropic’s request to block its designation as a ‘supply chain risk,’ pointing to the company’s insistence on restricting the use of its AI for autonomous weapons and domestic surveillance. The DoJ is arguing that the move is lawful, reasonable, and grounded in national security, not a violation of free speech.

Egypt. On 14 March 2026, Egypt published the National Guidelines for Trustworthy and Responsible AI. The Guidelines provide a national reference for the responsible development, deployment, and oversight of AI across public and private sectors, ensuring AI use is safe, ethical, and transparent while supporting innovation aligned with Egypt’s Vision 2030 and the National AI Strategy. Complementing the National AI Governance Framework, which defines what should be governed, these Guidelines specify how to comply, offering methodologies, metrics, and checklists to operationalise ethical principles. Targeted at data scientists, compliance officers, and developers, they provide actionable directions to protect individual rights, promote societal well-being, enhance accountability and transparency, and foster innovation grounded in safety. The Guidelines also align Egypt with international standards and engage government entities, private enterprises, and community actors in responsible AI governance.

The EU. Tensions are emerging in the EU over AI infrastructure investment, with France, Poland, Austria, and Lithuania pushing to reserve part of the €20 billion AI Gigafactory project for European technologies, while Germany is sceptical about linking the project to digital sovereignty goals. Meanwhile, Germany is pursuing a major expansion of domestic data centres and AI processing power, supported by regulatory reforms, tax incentives, and land allocation to attract investment, aiming to reduce reliance on foreign providers.In parallel, the EU is tightening AI regulations: the Council has endorsed proposals to ban AI from generating non-consensual sexual content (CSAM), adjust high-risk AI compliance timelines, and streamline the AI Act, including exemptions for some SMEs, registration requirements, and clarified oversight responsibilities. These moves reflect Europe’s broader effort to secure sovereign AI infrastructure and ensure safe, accountable AI deployment.

Fighting fraud: The Global Summit in Vienna

INTERPOL has launched a new global task force at the Global Fraud Summit 2026 as part of a more coordinated, data-driven response to the rapid global expansion of financial fraud.

The task force is jointly developed by the UK’s Home Office and INTERPOL and is codenamed Operation Shadow Storm. The task force will target scam centres and their links to cybercrime and human trafficking, using tools such as stop-payment mechanisms and international intelligence-sharing networks. The initial focus of the task force will be dismantling criminal operations across Southeast Asia.

Simultaneously, major technology and consumer-facing companies, including Google, Amazon, Meta, and OpenAI, have signed the ‘Industry Accord Against Online Scams and Fraud’ at the Global Fraud Summit 2026.

The companies pledged to focus on deploying proactive security measures and AI-driven detection systems; strengthening information sharing between industry and law enforcement to better identify and respond to fraud; enhancing resilience through advanced defensive technologies and rapid response mechanisms; and improving public education to help individuals recognise and avoid scams.

Zooming out. Online scams are growing in scale and sophistication, aided by AI-generated content and cross-platform operations. Data shows that consumers lost over $16 billion to online scams in 2024.

EU rules on CSAM detection lapse, leaving a regulatory gap

The EU has been unable to reach an agreement on extending temporary rules that allow online platforms to detect child sexual abuse material, leaving the current framework set to expire in April. The existing rules, in place since 2021, permit technology companies to voluntarily scan their services for harmful content, supporting efforts to identify and remove illegal material.

But negotiations between the European Parliament and member states stalled over key issues — especially whether such measures should apply to encrypted services.

What’s next? Attention now shifts to the long-delayed permanent framework (the Child Sexual Abuse Regulation).

Cybersecurity: On the defensive

As cyber capabilities are now deeply integrated into broader conflict dynamics, countries are increasingly deploying the full range of tools at their disposal.

The EU has imposed sanctions over cyber attacks targeting its member states and partners, listing China-based Integrity Technology Group and Anxun Information Technology, as well as Iran-based Emennet Pasargad, along with Anxun’s co-founders. Integrity Technology is assessed to have facilitated the compromise of over 65,000 devices across six member states. Anxun is assessed to have provided offensive cyber capabilities targeting critical infrastructure, and two of the company’s co-founders have been individually designated for their roles in these operations. Emennet is assessed to have a compromised digital advertising infrastructure to disseminate disinformation during the 2024 Paris Olympics. The sanctions entail an asset freeze and a travel ban for the listed individuals. The EU citizens and entities are additionally prohibited from making funds available to the designated companies.

Long constrained by a defensive security doctrine, Japan will introduce ‘hack-back’ powers from October. The change comes around as part of Japan’s ‘Active Cyber Defense’ law, which was passed in 2025 and is rolling out in incremental stages through 2027. The framework enables authorities to pre-emptively identify and neutralise hostile infrastructure, while also mandating incident reporting by critical infrastructure operators and strengthening coordination between the National Police Agency, intelligence services, and the Self-Defense Forces (SDF).

After last week’s cyberattack on US medical giant Stryker, the FBI seized four websites tied to Handala, the pro-Iranian hacking group that claimed responsibility for the attack, and to Iran’s Ministry of Intelligence and Security (MOIS). The sites were used to claim responsibility for cyberattacks, leak stolen data, and incite violence against journalists, dissidents, and Israeli individuals, investigators claim. Investigators found the domains were interconnected through shared infrastructure and a coordinated operational playbook involving disruptive cyberattacks and ‘faketivist’ propaganda. Handala posted a message stating that ‘This act of digital aggression only serves to highlight the fear and anxiety our actions have instilled in the hearts of those who oppress and deceive. They may have taken down our website, but they will never take down our spirit, our resolve, or the power of truth.’

CISA published an alert urging organisations to harden end point management system configurations to defend against similar malicious activity.
A reminder that last week, Iran’s semi-official Tasnim News Agency, which is linked to the country’s Islamic Revolutionary Guard Corps, listed a number of major US tech companies as potential targets: Google, Microsoft, Palantir, IBM, Nvidia and Oracle.

China’s five-year plan to lead in tech

A new five-year development plan approved by lawmakers in Beijing places innovation and advanced technology at the centre of future economic growth, with the explicit aim of strengthening technological capabilities and positioning China as a leading global tech power.

The strategy sets out ambitions to upgrade the industrial sector, expand domestic research capacity, and reduce reliance on foreign technologies. Priority areas include AI, robotics, aerospace, biotechnology, and quantum computing

China plans to expand AI-related industries, invest in large-scale computing infrastructure, and support the development of advanced systems capable of performing complex tasks beyond traditional chatbot applications.

At the same time, China is set to scale up spending on science and technology, with government research budgets projected to grow by around 10% annually.

The plan also targets an increase in overall R&D investment of at least 7% per year.

The big picture. This strategy is both a response to external pressures and a long-term shift toward higher-value, tech-driven economic development. China aims to end its reliance on foreign innovation and directly challenge Western dominance in critical future industries, such as AI and quantum computing, a goal that is certainly influenced by continued tensions with the USA over trade and technology restrictions.

Brazil’s ECA Digital goes into force

Brazil has started enforcing a new law aimed at strengthening protections for children online, marking a significant shift in how digital platforms are regulated in the country. The legislation, known as ECA Digital, introduces stricter rules for technology companies and will test whether stronger oversight can translate into real-world impact.

The law, which takes effect this week, allows authorities to impose warnings and fines of up to $10 million for violations. In severe cases, courts may order the suspension or banning of platforms operating in Brazil. The measure was passed rapidly following public outrage over online content involving the sexualisation of minors.

ECA Digital builds on Brazil’s existing child protection framework and adapts it to the digital environment. It introduces obligations such as age verification, stricter content moderation, and mechanisms to remove harmful material involving minors without requiring a court order.

The law also targets platform design, requiring companies to limit features that may encourage compulsive use among children. This includes restrictions on excessive notifications, profiling for targeted advertising, and design elements that prolong user engagement.

What’s next? Enforcement of ECA Digital will be led by Brazil’s data protection authority, ANPD, alongside a new screening centre within the Federal Police. However, implementation challenges remain, including limited regulatory capacity and the short timeline between the law’s approval and enforcement.

View all updates

LOOKING AHEAD

The Geneva Graduate Institute is organising a briefing lunch on 23 March to examine evolving transatlantic dynamics at the intersection of US politics and the global influence of major technology platforms. The discussion will explore how recent political developments in the USA and the concentration of technological power shape Europe’s position, including questions of dependency, regulation, and strategic autonomy.

The International Labour Organization (ILO) is hosting a session on the macroeconomic impacts of AI on 25 March, showcasing a new World Bank Group model that treats AI as a structural transformation of production. The tool simulates how AI adoption affects sectors, occupations, and prices, helping policymakers assess implications for growth, equity, and structural change. A first case study in Poland will explore its application, with potential use in other emerging and middle-income economies.

The 14th Ministerial Conference of the World Trade Organization (MC14) is scheduled to take place from 26 to 29 March 2026 at the Palais des Congrès in Yaoundé, Cameroon. The Ministerial Conference, convening biennially, holds the highest authority within the WTO. It brings together all WTO members, comprising countries or customs unions, enabling decisions on various issues covered by the multilateral trade agreements. Members are expected to seek ministerial endorsement of a structured work plan, and preparatory breakout sessions on reform are included in the conference roadmap. These efforts reflect the central role that reform is anticipated to play in Yaoundé and the intention to initiate high-level political exchanges among ministers on this topic during the conference.

READING CORNER

ZqSAvFA2 BLOG featured image 2026 34 Trumps Cyber Strategy for America

Trump’s Cyber Strategy for America: Implications for international cyber norms and diplomacy – Diplo

The Trump Administration’s 2026 Cyber Strategy signals a shift from rule-based cyber governance to a power-driven approach centred on offensive capabilities, private-sector mobilisation, and transactional diplomacy. Diplo experts examine what this posture means for international cyber norms, multilateral processes, and the future of cyber diplomacy.

AI Regulation meets enforcement reality: How the rules actually work – Diplo

How are the EU AI Act and California’s new laws actually being enforced in 2026? We look at the “AI traffic cops” and what happens when rules are broken.

Code of opacity: How AI is enabling corruption – Diplo

Policymakers worldwide are caught between awe and apprehension over AI. They recognise its potential to accelerate productivity and scientific progress while worrying about threats to jobs, human rights, and social cohesion. Yet they’re missing a critical risk: AI is becoming a code of opacity within government. Without adequate oversight, AI systems can facilitate corruption—eroding public trust in both the technology and the institutions deploying it.

Weekly #253 Measuring hate’s footprint

6 March – 13 March 2026

HIGHLIGHT OF THE WEEK

Measuring hate’s footprint

Spain has unveiled HODIO, a new digital tool designed to systematically measure and expose the prevalence of hate speech across social media platforms. The tool—short for Huella del Odio y Polarización (Footprint of Hatred and Polarisation)—was announced by Pedro Sánchez at the inaugural Forum Against Hate in Madrid. It will be managed by the Spanish Observatory on Racism and Xenophobia (OBERAXE).

How will it work? HODIO will combine quantitative analysis, AI tools, and expert review to assess the scale and patterns of hate speech on major platforms. Every six months, the observatory will publish a report ranking platforms according to users’ exposure to hostile messages, allowing comparisons across services.

The goal. By measuring the ‘footprint’ of hate, the government hopes to create stronger evidence for policymaking and increase pressure on platforms to take action against harmful content.

Person, Book, Comics, Publication, Animal, Dinosaur, Reptile, Face, Head

Why does it matter? The tool reflects a growing push by governments to gather data on online harms and better understand how hate speech spreads across platforms. For instance, France is investigating influencer-driven hate and platform policies, Brazil is using AI to monitor anti-LGBTQ online content, and Indonesia has created a real-time dashboard revealing stark differences in platform toxicity during elections.

But: a slippery slope. Critics have raised concerns about the transparency of HODIO and how authorities will define and classify hate speech, warning that poorly defined criteria could infringe on freedom of expression.

Why does it matter? Ultimately, the dispute exposed a deeper structural tension. Advanced AI systems are increasingly central to military planning, logistics, intelligence analysis, and battlefield decision-making. At the same time, leading AI firms have articulated ethical boundaries around surveillance, lethal autonomy, and dual-use risks.

The confrontation between Anthropic and the Pentagon crystallised the question of who determines those boundaries when national security and corporate governance collide.

IN OTHER NEWS LAST WEEK

This week in AI governance

The USA. The US government is facing two lawsuits from AI firm Anthropic after the Pentagon designated the company a supply-chain risk, effectively barring its technology from defence contracts. Anthropic argues the move is unlawful and politically motivated, claiming the government is retaliating against the company for refusing to allow its AI models to be used for domestic surveillance or fully autonomous weapons. The lawsuits, filed in courts in California and Washington, D.C., challenge the rare use of national-security supply-chain rules against a US technology company.

The legal dispute has drawn support from across the tech sector, with companies including Microsoft, Google, Amazon and OpenAI backing Anthropic’s legal challenge through amicus filings. Industry leaders warn that the government’s designation could set a precedent that destabilises the US AI ecosystem and disrupts suppliers working with both government and private-sector AI systems.

The EU. The European Commission has released a second draft of its Code of Practice on marking and labelling AI-generated content, part of efforts to help companies comply with transparency requirements under Article 50 of the EU Artificial Intelligence Act. Section 1 of the code focuses on providers of generative AI systems and proposes a multi-layered approach to marking AI-generated content, including digitally signed metadata, imperceptible watermarking, and optional fingerprinting or logging. Providers are also expected to make detection tools available so users and authorities can verify whether content was generated or manipulated by AI. Section 2 addresses deployers of AI systems, requiring clear disclosure when deepfakes or AI-generated text intended to inform the public have been artificially generated or manipulated, using visible and accessible labels.

Meta reopens WhatsApp to third-party AI Chatbots amid EU pressure

Meta has announced that third-party AI chatbots will once again be allowed to operate through WhatsApp in Europe for a fee, reversing earlier restrictions that limited access to rival chatbot services on the platform.

The decision follows pressure from the European Commission, which had warned it could impose interim competition measures. Under the new arrangement, companies will be able to distribute general-purpose AI chatbots via the WhatsApp Business API for 12 months. The change is intended to give European regulators time to complete their investigation while allowing competing AI services to operate within the platform ecosystem.

What’s next? EU competition chief Teresa Ribera said that the Commission would examine Meta’s proposal and decide whether further intervention is necessary.

UK rejects social media ban, opting for flexible measures

A proposed social media ban for under-16s has been rejected by UK MPs, with 307 voting against and 173 in favour. Instead, the parliamentarians supported an alternative plan put forward by Education Minister Olivia Bailey to give ministers flexible powers, enforceable after a consultation on online safety concludes.

Under this plan, technology secretary Liz Kendall could ‘restrict or ban children of certain ages from accessing social media services and chatbots’. She could also limit children’s VPN use, restrict access to addictive features, and change the age of digital consent in the UK.

At the same time, UK online safety regulators Ofcom and the Information Commissioner’s Office have also called on social media firms to better protect children online. Ofcom issued an open letter urging tech firms to keep underage children off their platforms, emphasising the importance of robust age verification systems. The guidance is backed by the Information Commissioner’s Office (ICO), which highlighted the need to protect children’s personal data and strengthen compliance with existing regulations.

Looking ahead. Where Bailey’s plan goes depends on the online safety consultation, which is set to end on 26 May. A joint statement by Ofcom and ICO, expected in March 2026, will clarify how online safety and data protection intersect in the context of age assurance, offering updated guidance to platforms on their responsibilities.

Why does it matter? In an era when an increasing number of countries are considering social media bans for minors, the UK’s recent decision stands out for explicitly rejecting a total ban. Its approach could set an influential example, potentially leading some governments that had been weighing bans to rethink their plans.

Cybersecurity: On the offensive

President Donald Trump released his administration’s national cybersecurity strategy, outlining priorities across six policy areas: offensive and defensive cyber operations, federal network security, critical infrastructure protection, regulatory reform, emerging technology leadership (including in AI), and workforce development.

Trump also signed an executive order the same day, directing the attorney general to prioritise cybercrime prosecution, tasking agencies with reviewing tools to counter international criminal organisations, and assigning the Department of Homeland Security expanded training responsibilities.

The strategy document spans five pages of substantive text, with administration officials describing it as intentionally high-level. The White House stated that more detailed implementation guidance would follow.

In the pipeline. A full analysis of the strategy by Diplo’s cybersecurity policy team. Stay tuned!

On the heels of the strategy comes the latest cyber episode in the USA-Israel-Iran conflict, with pro-Iranian hacker group Handala claiming responsibility for a cyberattack on US medical device giant Stryker. The group has stated that the cyberattack is retaliation for a missile strike on an elementary school in Iran. Stryker confirmed the cyberattack in a statement, noting that order processing, manufacturing and shipping are disrupted, but that connected products have not been impacted.

Meanwhile, European intelligence agencies are warning of a growing cyber-espionage campaign targeting accounts on encrypted messaging platforms such as Signal and WhatsApp. Authorities in the Netherlands reported that hackers—believed to be linked to Russia—have launched large-scale phishing operations aimed at diplomats, military personnel, government officials, and journalists. Instead of breaking the apps’ encryption, attackers trick users into sharing verification codes or linking devices, allowing them to take over accounts and access sensitive conversations.

Portugal’s intelligence service has issued a similar alert, describing a global campaign by foreign state-backed actors seeking access to the messaging accounts of officials and others with privileged information. Once inside an account, attackers can read chats, access shared files, and use the compromised profile to target additional victims through further phishing attempts.

The big picture. The past week confirms that cyberspace has become a domain of unrelenting offensive action. With the Global Mechanism on ICT security set to begin its work at the end of March, how these very incidents will inform its discussions is the question.

View all updates

LOOKING AHEAD

The Global Fraud Summit 2026, convened by the UNODC in cooperation with INTERPOL, will take place on 16–17 March at the Vienna International Centre, Austria. This ministerial-level meeting brings together government officials, law enforcement, international organisations, the private sector, civil society, and academics to tackle fraud as a transnational threat. Participants will explore emerging trends, digital and cross-border challenges, prevention strategies, enforcement measures, and international information-sharing mechanisms.

WIPO will launch the AI Infrastructure Interchange (AIII) on 17 March in Geneva and online. The programme includes keynote remarks, panel discussions, and presentations addressing the role of technical collaboration between creators, rightsholders, and technology companies. Participants will also discuss the objectives of the AIII initiative and the establishment of a Technical Exchange Network intended to support ongoing expert dialogue on practical challenges and opportunities. Registration for the event is open.

Diplo is organising a webinar, ‘Technology Innovations for Creative Diplomacy’, on 18 March to examine the impact of technological innovation on diplomatic practice. The webinar will showcase creative applications of digital tools in public diplomacy and international engagement, and share practical experiences from technology-driven initiatives. Registration for the event is open.

READING CORNER

Automated exclusion: The crisis of inaccessible AI for persons with disabilities – Diplo

In the global rush to regulate AI, a consensus seems to have formed: that the real drama of AI lies in the future – in existential risks, deepfakes, or algorithmic bias. But for persons with disabilities, the crisis of AI is already here, and it is older than the technology itself. It is the story of being told a system is efficient and modern, only to find the door locked from the inside, writes Muhammad Shabbir.

After universality: Ubuntu, AI governance and the problem of moral monoculture – Diplo

Can AI ethics truly be universal? Or are we quietly building a global moral monoculture into our machines? Emmanuel Elolo Agbenonwossi explores how ideas like Ubuntu challenge dominant AI governance narratives and why pluralism may matter more than universality.

Digital ghosts – Diplo

The management of a digital afterlife involves more than just closing accounts. It concerns the long-term security and ethical handling of personal data. Slobodan Kovrlija examines the growing need for international standards for digital legacy and what this means for the future of digital rights.

Lettre d’information du Digital Watch – Numéro 107 – Mensuelle février 2026

Rétrospective de février 2026

La sécurité en ligne des enfants est sous les projecteurs. Le procès de Los Angeles sur l’addiction aux réseaux sociaux et le procès sur la protection des consommateurs à Santa Fe ont débuté, ce qui est une première pour les plateformes de réseaux sociaux qui doivent se défendre devant un tribunal. Parallèlement, de plus en plus de pays envisagent d’interdire l’accès aux réseaux sociaux pour les enfants.

Les points forts de ce mois :

Pourquoi le cyberespace n’existe pas : 30 ans après la Déclaration d’indépendance du cyberespace, nous examinons comment le mythe du cyberespace en tant que monde à part entière façonne aujourd’hui la gouvernance de l’IA.

Anthropic contre le Pentagone : L’entreprise d’IA Anthropic a été exclue des travaux avec le Pentagone après avoir refusé de renoncer à ses garanties éthiques, ce qui a donné lieu à un affrontement majeur entre la Silicon Valley et l’armée.

Souveraineté technologique : de l’autonomie stratégique européenne à la riposte américaine, la souveraineté numérique continue de dominer l’année 2026.

En route vers Genève 2027 : dix étapes clés avant le prochain Sommet mondial sur l’IA, développé par les recherches de DiploAI.

En plus : les principales évolutions des politiques numériques en février et un récapitulatif de l’actualité à Genève.

Crowd, Person, People, Audience, Baby, Face, Head, Speech, Debate

Coup d’œil : les événements qui ont marqué le mois de février

TECHOLOGIES

Aux Pays-Bas, une cour d’appel néerlandaise a ordonné une enquête approfondie sur Nexperia et a confirmé les décisions prises précédemment de suspendre son ancien PDG, lié à la société mère chinoise Wingtech. La décision de la cour renforce la surveillance de la gouvernance et des activités de l’entreprise, reflétant les préoccupations plus générales de l’Europe concernant la propriété et le contrôle étrangers d’actifs stratégiques sensibles dans le domaine des semi-conducteurs, en particulier pour les secteurs de l’automobile et de la technologie. Nexperia s’est engagée à coopérer pleinement à l’enquête, tandis que Wingtech a critiqué cette décision, la jugeant préjudiciable à l’industrie mondiale.

Taiwan Semiconductor Manufacturing Company (TSMC) a annoncé qu’elle produirait des puces IA avancées de 3 nanomètres dans sa deuxième usine de fabrication à Kumamoto, au Japon. Cette expansion souligne la demande mondiale croissante en processeurs IA et s’aligne sur la stratégie du Japon visant à renforcer les capacités nationales en matière de semi-conducteurs et à diversifier la production critique en dehors des pôles traditionnels.

Le Royaume-Uni et la Bulgarie ont convenu d’approfondir leur coopération dans le domaine des semi-conducteurs. Le Réseau britannique pour la science et la technologie et le ministère britannique des Affaires et du Commerce établissant un lien entre l’expertise britannique en matière de semi-conducteurs composés et la base manufacturière bulgare. Le principal résultat de ce partenariat est l’avancement du projet de construction d’une usine de plaquettes de « carbure de silicium vert » d’une valeur de 350 millions d’euros en Bulgarie, ainsi que de nouveaux partenariats en matière de R&D et d’industrie, notamment un protocole d’accord de recherche entre les universités de Glasgow et de Sofia et un accord de coopération industrielle avec le groupe allemand Wacker Chemie AG.

Un examen de sécurité nationale américain retarde l’octroi des licences permettant à Nvidia d’expédier ses puces IA H200 en Chine, empêchant ainsi les clients chinois de passer ou de confirmer leurs commandes près de deux mois après que la Maison Blanche ait donné son accord de principe. Bien que le département du Commerce ait terminé son évaluation, les consultations interinstitutionnelles, notamment entre les départements d’État, de la Défense et de l’Énergie, sont toujours en cours pour négocier des garanties supplémentaires et des conditions potentielles telles que la répartition des expéditions, les tests et les rapports sur l’utilisation finale. Ces retards perturbent déjà les prévisions de la demande et la planification de la production tout au long de la chaîne d’approvisionnement de Nvidia et poussent les entreprises chinoises à explorer d’autres options. Défense et Énergie, sont toujours en cours de négociation concernant des garanties supplémentaires et des conditions potentielles telles que la répartition des expéditions, les tests et les rapports sur l’utilisation finale. Ces retards perturbent déjà les prévisions de la demande et la planification de la production tout au long de la chaîne d’approvisionnement de Nvidia et poussent les entreprises chinoises à explorer d’autres moyens de s’approvisionner en puces IA.

Découvrez les développements de ce mois-ci en matière de gouvernance de l’IA dans notre section dédiée à la newsletter sur l’IA.

INFRASTRUCTURE

Les ingénieurs procèdent actuellement à la récupération du TAT-8, le premier câble transatlantique à fibre optique, des fonds marins de l’Atlantique, plus de trois décennies après qu’il ait révolutionné les communications mondiales. Bien qu’il ait été mis hors service en 2002 à la suite d’une défaillance irréparable, le câble est resté immergé jusqu’à présent. Cette opération permet de dégager les fonds marins pour y installer de nouvelles infrastructures et de récupérer les composants en fibre de verre, en cuivre et en acier à des fins de recyclage, dans un contexte de pénurie mondiale de métaux.

Starlink, de SpaceX, a obtenu l’autorisation réglementaire d’opérer au Vietnam, élargissant ainsi sa présence mondiale et renforçant son rôle de fournisseur de connectivité alternatif sur des marchés fortement réglementés. Parallèlement, un responsable russe a reconnu que les systèmes Starlink avaient été hors service pendant deux semaines dans certaines régions de Russie.

Une coalition de grandes entreprises technologiques a annoncé la création de la Trusted Tech Alliance (TTA) et présenté cinq principes fondamentaux pour définir ce qui constitue une infrastructure numérique « fiable » : une gouvernance d’entreprise transparente, un développement sécurisé et une évaluation indépendante, une supervision de la chaîne d’approvisionnement, une surveillance de l’écosystème et une responsabilité accrue des entreprises.

CYBERSÉCURITÉ

Le ministère de la Sécurité publique de la République populaire de Chine a rédigé un projet de loi autorisant les autorités à imposer des interdictions de sortie du territoire pouvant aller jusqu’à trois ans aux cybercriminels condamnés et à ceux qui soutiennent ou facilitent de telles activités. Ce projet de loi interdirait également l’entrée sur le territoire aux contrevenants, étendrait la juridiction aux ressortissants chinois à l’étranger, ciblerait les entités étrangères jugées préjudiciables aux intérêts nationaux et renforcerait les contrôles sur les contenus en ligne jugés faux ou perturbateurs. Cette proposition pourrait avoir des répercussions sur les entreprises mondiales, la coopération transfrontalière et la mobilité internationale des professionnels de la technologie.

Les États-Unis et Israël ont lancé des frappes militaires coordonnées contre l’Iran, accompagnées d’opérations cybernétiques menées par l’USCYBERCOM. Ces opérations cybernétiques, associées à des opérations spatiales, ont perturbé les communications iraniennes. La riposte cybernétique de l’Iran s’est jusqu’à présent limitée à des attaques par déni de service distribué, à l’usurpation de signaux GPS près du détroit d’Ormuz et au piratage de caméras IP pour soutenir les opérations de missiles, ainsi qu’à une coupure Internet à l’échelle nationale.

La Commission européenne a lancé ProtectEU, un nouveau programme de lutte contre le terrorisme qui renforce la réponse de l’Union européenne aux menaces en constante évolution, particulièrement celles amplifiées par les outils numériques, en renforçant l’analyse du renseignement et le soutien d’Europol, en resserrant la coopération avec les plateformes afin de supprimer plus rapidement les contenus extrémistes et en renforçant l’application de la loi sur les services numériques, tout en proposant un cadre européen de réponse aux crises en ligne afin de coordonner l’action des entreprises technologiques lors d’incidents de sécurité et en élargissant les mesures visant à protéger les espaces publics, les infrastructures critiques et à perturber le financement du terrorisme, y compris via les crypto-actifs.

Découvrez les dernières évolutions en matière de sécurité des enfants dans notre rubrique qui leur est dédiée.

ÉCONOMIE

La Commission européenne et la Commission de la concurrence et de la consommation du COMESA mènent chacune de leur côté une enquête sur les craintes que Meta abuse de sa position dominante en limitant l’accès des assistants IA tiers à WhatsApp tout en privilégiant sa propre IA Meta. La Commission européenne a déjà officiellement notifié à Meta qu’elle avait enfreint le droit européen de la concurrence et envisage des mesures provisoires pour empêcher la poursuite de cette exclusion et protéger l’entrée de concurrents sur le marché.

La France durcit sa position à l’égard des détaillants en ligne à très bas prix, le ministre Serge Papin ayant déclaré 2026 « année de la résistance » aux plateformes telles que Shein. Le gouvernement fait valoir que les places de marché mondiales bénéficient de normes réglementaires moins strictes que les magasins physiques français. Paris fait appel d’une décision de justice qui a autorisé Shein à poursuivre ses activités malgré des produits inappropriés et prépare une législation permettant aux autorités de suspendre les plateformes en ligne sans autorisation judiciaire préalable, élargissant ainsi les pouvoirs exécutifs sur l’économie numérique.

Le parti AK au pouvoir en Turquie a présenté un projet de loi visant à officialiser la taxation des cryptomonnaies en liant les règles relatives aux actifs numériques à la loi sur les marchés financiers et en exigeant des plateformes agréées qu’elles retiennent une taxe de 10 % sur les gains et les revenus liés aux cryptomonnaies chaque trimestre pour les particuliers et les entreprises, y compris les résidents et les non-résidents. La proposition ajoute également une taxe de transaction de 0,03 % sur les prestataires de services cryptographiques, oblige les investisseurs utilisant des plateformes non agréées à déclarer leurs gains chaque année et permettrait au président d’ajuster le taux de retenue de 0 % à 20 % en fonction de facteurs tels que le type de jeton ou la période de détention. Le nouveau régime fiscal devrait entrer en vigueur deux mois après sa publication si le projet de loi est adopté.

Le ministre sud-coréen des Finances, Koo Yun-cheol, s’est engagé à réformer de toute urgence la manière dont les agences gouvernementales traitent les cryptomonnaies saisies et détenues par l’État après plusieurs défaillances en matière de conservation, notamment un cas dans lequel la police de Séoul aurait perdu l’accès à 22 BTC (environ 1,4 million de dollars) lorsque les clés privées n’ont pas été correctement conservées et qu’un tiers a été autorisé à gérer les actifs. Les procureurs enquêtent également sur des allégations de corruption liées à cet incident, et le ministère des Finances affirme que le gouvernement ne détient des cryptomonnaies que dans le cadre de mesures coercitives légales, telles que des saisies dans des affaires fiscales et pénales.

La banque centrale russe indique qu’elle intensifie sa lutte contre les systèmes pyramidaux basés sur les cryptomonnaies, signalant que deux tiers de ces opérateurs s’appuient désormais sur les cryptomonnaies et que les fonds des victimes ont été transférés vers plus de 4 600 portefeuilles contrôlés par des fraudeurs en 2025. L’autorité de régulation indique avoir identifié 7 087 escroqueries en ligne l’année dernière, bloqué 21 500 pages web et publications sur les réseaux sociaux liées à des escroqueries, et exhorte les Russes à ne recourir qu’à des prestataires d’investissement agréés, alors que les autorités renforcent la surveillance des fraudes en ligne propagées via les réseaux sociaux, les applications de chat et les appels téléphoniques.

DROITS DE L’HOMME

L’UE a abandonné son projet de révision de la définition des « données à caractère personnel » dans le projet de paquet omnibus du RGPD, suite à une forte opposition des régulateurs nationaux et de la société civile, et a choisi de maintenir intact le champ d’application actuel du règlement. L’attention se porte désormais sur les prochaines orientations du Comité européen de la protection des données en matière de pseudonymisation, qui devraient clarifier la manière dont les principales garanties doivent être appliquées dans la pratique, ce qui témoigne d’une préférence générale pour la clarté réglementaire et les orientations de mise en œuvre plutôt que pour la réouverture des concepts fondamentaux de la législation en matière de protection de la vie privée.

Les négociations entre l’Australie et les États-Unis sur l’élargissement du partage des données biométriques ont suscité l’inquiétude des défenseurs de la vie privée et des commentateurs juridiques. Selon certaines informations, ces discussions pourraient élargir l’accès des États-Unis aux données biométriques sensibles australiennes, notamment les images faciales, les empreintes digitales et les données d’identité, en contournant les cadres traditionnels de coopération juridique au cas par cas.

L’autorité italienne de protection de la vie privée a ordonné à Amazon Italia Logistics de cesser le traitement des données sensibles des employés sur son site de Passo Corese et d’arrêter l’utilisation des données collectées par les caméras de surveillance installées près des toilettes et des zones de pause, après avoir découvert que l’entreprise enregistrait des informations telles que l’état de santé des travailleurs, les activités syndicales/de grève et des détails privés sur leur famille, et les conservait pendant une durée pouvant aller jusqu’à 10 ans, ce qui dépasse largement ce que les autorités considèrent comme légal pour la gestion du lieu de travail.

La Commission nigériane de protection des données (NDPC) a ouvert une enquête sur le géant chinois du commerce électronique Temu pour violation présumée de la loi nigériane sur la protection des données. Les autorités enquêtent sur les pratiques de l’entreprise en matière de traitement des données, en particulier sur des allégations de traitement non transparent des données, de mécanismes de surveillance intrusifs, de transferts transfrontaliers et de non-respect éventuel des limites imposées à la collecte de données. Temu s’est engagé à coopérer, les régulateurs avertissant que le non-respect de la réglementation pourrait entraîner des sanctions légales et créer un précédent en matière de gouvernance des données sur le plus grand marché numérique d’Afrique.

Un projet de loi sur la souveraineté cognitive et la protection de l’attention humaine a été présenté à la Chambre des députés argentine. Il propose la mise en place d’un cadre réglementaire reconnaissant l’autonomie cognitive comme un bien protégé par la loi en vertu du droit constitutionnel et du droit international des droits de l’homme, exigeant la transparence, la personnalisation opt-in et des alternatives non algorithmiques sur les plateformes à large audience. Il suggère également d’imposer des paramètres stricts « sécurisés par défaut » pour les mineurs, d’interdire le profilage comportemental et la publicité ciblée pour les utilisateurs de moins de 13 ans, et d’imposer des évaluations d’impact, des obligations d’enregistrement, des audits et des rapports annuels de transparence.

SOCIOCULTUREL

L’UNESCO et l’université Hamad Bin Khalifa (HBKU) ont lancé une chaire UNESCO sur les technologies numériques et le comportement humain au Qatar afin d’étudier l’impact des technologies émergentes sur la vie quotidienne, en mettant l’accent sur le bien-être numérique, la conception éthique et des environnements en ligne plus sains. Le programme abordera des questions telles que la dépendance à Internet, la cyberintimidation et la désinformation, et vise à établir un lien entre la recherche et le dialogue politique entre les gouvernements, les organisations internationales et les universités afin de promouvoir un développement technologique plus responsable.

Le Royaume-Uni introduit une législation obligeant les entreprises technologiques à supprimer les images intimes non consensuelles dans les 48 heures suivant leur signalement. En vertu du projet de loi actualisé sur la criminalité et la police, les entreprises qui ne se conforment pas à cette obligation s’exposent à des amendes pouvant atteindre 10 % de leur chiffre d’affaires mondial ou à des restrictions de service potentielles, dont l’application est supervisée par l’Ofcom.

Le département d’État américain serait en train de préparer le lancement de « freedom.gov », un portail en ligne conçu pour aider les utilisateurs du monde entier, y compris en Europe et ailleurs, à contourner les restrictions locales en matière de contenu et à accéder à des informations bloquées, notamment celles que leurs gouvernements classent comme discours haineux ou propagande terroriste.

L’UE enquête pour déterminer si les éléments de conception de Shein, tels que l’engagement ludique et les algorithmes de recommandation opaques, compromettent la sécurité des consommateurs et les obligations de transparence prévues par le DSA. Les enquêteurs examineront si Shein a manqué à son obligation d’empêcher la vente de produits illégaux, notamment des articles pouvant constituer du matériel pédopornographique. Les systèmes d’atténuation des risques, les processus de retrait des produits et le respect des exigences en matière d’options de recommandation sans profilage de Shein seront évalués, avec des amendes potentielles pouvant atteindre 6 % du chiffre d’affaires mondial en cas de violations confirmées.

La Commission européenne a conclu à titre préliminaire que la conception de TikTok enfreint la loi sur les services numériques (DSA) de l’Union européenne en raison de fonctionnalités que la Commission considère comme addictives, telles que le défilement infini, la lecture automatique, les notifications push et son système de recommandation hautement personnalisé. Selon la Commission, les mesures de protection existantes sur TikTok, telles que la gestion du temps d’écran et les outils de contrôle parental, ne semblent pas suffisantes pour atténuer les risques associés à ces choix de conception. À ce stade, la Commission indique que TikTok devrait modifier la conception fondamentale de son service. Les mesures envisageables comprennent la suppression progressive ou la limitation du défilement infini, l’introduction de pauses plus efficaces dans le temps d’écran, y compris la nuit, et l’ajustement de son système de recommandation afin de réduire les effets addictifs.

DÉVELOPPEMENT

La Malaisie a imposé une interdiction immédiate et totale de toutes les importations de déchets électroniques, reclassant ces derniers dans la catégorie « interdiction absolue » de ses règles d’importation, à la suite d’une enquête approfondie sur la corruption dans la surveillance de ce secteur. Les autorités affirment que cette mesure vise à mettre fin au dumping étranger et à protéger la santé publique et la sécurité nationale, tout en avertissant que les contrôles seront renforcés pour empêcher la contrebande. L’enquête aurait conduit à la détention de hauts responsables environnementaux et au gel de leurs avoirs.

Le Gabon a imposé une suspension indéfinie des plateformes de réseaux sociaux, invoquant la propagation de fausses informations, le cyberharcèlement et la divulgation non autorisée de données personnelles. L’autorité de régulation des médias du Gabon, la Haute Autorité de la communication (HAC), a déclaré que les mesures de modération existantes ne fonctionnaient pas et que la fermeture était nécessaire pour mettre fin aux violations du Code des communications de 2016 du Gabon.

La Commission européenne a proposé d’ouvrir des négociations afin d’intégrer l’Albanie, la Bosnie-Herzégovine, le Kosovo*, le Monténégro, la Macédoine du Nord et la Serbie dans le régime « Roam Like at Home » (Itinérance au prix national) de l’UE. Si cette mesure est mise en œuvre, les citoyens et les entreprises pourront passer des appels, envoyer des SMS et utiliser des données mobiles à l’étranger aux tarifs nationaux, tant lorsqu’ils se rendent dans l’UE que lorsque les citoyens de l’UE voyagent dans la région.La Commission a adopté des propositions de mandats de négociation et sollicite actuellement l’approbation du Conseil européen pour entamer des discussions officielles. Le gouvernement du Cap-Vert a lancé Gov.CV, un portail numérique unifié conçu pour centraliser les services publics et rationaliser les interactions entre l’État, les citoyens et les entreprises. En regroupant les services, le gouvernement espère réduire les délais de traitement, limiter les redondances et offrir une expérience utilisateur plus transparente.

Gouvernance de l’IA en février

GOUVERNANCE MONDIALE

L’ONU. L’Assemblée générale a approuvé la création d’un organe consultatif scientifique mondial historique sur l’IA, le Comité scientifique international indépendant sur l’intelligence artificielle (IA), chargé de fournir des évaluations indépendantes et fondées sur des preuves concernant les technologies, les risques, les opportunités et les impacts de l’IA.

Premier du genre, le groupe d’experts a pour principale mission de « publier des évaluations scientifiques fondées sur des données probantes, synthétisant et analysant les recherches existantes relatives aux opportunités, aux risques et aux impacts de l’IA », sous la forme d’un « rapport de synthèse annuel pertinent sur le plan politique mais non prescriptif » qui sera présenté au Dialogue mondial sur la gouvernance de l’IA. Le groupe d’experts « fera également le point sur ses travaux jusqu’à deux fois par an afin de recueillir les avis dans le cadre d’un dialogue interactif entre l’Assemblée générale en séance plénière et les coprésidents du groupe ». Lors de la réunion inaugurale du groupe d’experts, M. Guterres a déclaré aux experts qu’ils avaient la lourde responsabilité de contribuer à définir la manière dont cette technologie serait utilisée « au profit de l’humanité ».

Sommet sur l’impact de l’IA 2026 à New Delhi. Pour la première fois dans les pays du Sud, l’Inde a accueilli le plus grand sommet mondial sur l’IA à Bharat Mandapam.

Les engagements de New Delhi Frontier AI Impact ont été dévoilés lors de l’ouverture du sommet, axés sur deux priorités fondamentales. La première, qui consiste à faire progresser la compréhension de l’utilisation de l’IA dans le monde réel, vise à générer des données anonymisées et agrégées afin d’éclairer l’élaboration des politiques sur l’impact de l’IA sur l’emploi, les compétences et la productivité. L’objectif est de soutenir une réglementation et une planification économique fondées sur des données probantes à mesure que l’adoption de l’IA s’accélère.

La seconde priorité, qui consiste à renforcer les évaluations multilingues et contextuelles, vise à améliorer les performances de l’IA dans les langues et les contextes culturels sous-représentés. Les organisations participantes collaboreront avec les gouvernements et les écosystèmes locaux pour développer des ensembles de données, des benchmarks et une expertise en matière d’évaluation, en mettant particulièrement l’accent sur les pays du Sud.

Au cœur du sommet, le Premier ministre Narendra Modi a dévoilé la « MANAV Vision », une approche de la gouvernance de l’IA centrée sur l’humain. Présentée comme une série de principes visant à placer l’humain au centre du développement et du déploiement de l’IA, MANAV signifie :

Moral and ethical systems (systèmes moraux et éthiques) – garantir que l’IA soit guidée par des normes éthiques
Accountable governance (gouvernance responsable) – règles transparentes et mécanismes de contrôle
National sovereignty (souveraineté nationale) – droits sur les données et les actifs numériques
Accessible and inclusive AI (IA accessible et inclusive) – éviter les monopoles et élargir la participation
Valid and legitimate systems (systèmes valides et légitimes) — technologies légales et vérifiables.

Modi a décrit ce cadre comme essentiel pour prévenir de futures disparités dans l’impact de l’IA et garantir que la technologie serve le bien-être de l’humanité. Il a également souligné que l’IA devrait être un moyen d’inclusion et d’autonomisation, en particulier pour les pays du Sud, plutôt qu’un outil qui concentre le pouvoir entre les mains de quelques acteurs.

L’édition 2027 du sommet sera accueillie par la Suisse. Pour en savoir plus sur la route vers le sommet de 2027, veuillez consulter notre newsletter dédiée.

INVESTISSEMENTS ET PROGRAMMES NATIONAUX

Etats-Unis. Sept géants technologiques (Google, Meta, Microsoft, Oracle, OpenAI, Amazon et xAI) ont signé l’engagement de la Maison Blanche en faveur de la protection des contribuables, encourageant les grandes entreprises technologiques à couvrir les coûts supplémentaires d’électricité liés à leur infrastructure d’IA et, dans certains cas, à investir dans la production d’énergie dédiée plutôt que de dépendre uniquement du réseau public.

Allemagne. L’Allemagne a présenté son projet de « Sovereign AI Factory », une initiative soutenue par le gouvernement visant à développer des modèles et des infrastructures d’IA souverains adaptés à la langue, au contexte culturel et aux besoins industriels locaux. Ce projet soutiendra l’innovation nationale en fournissant des ressources informatiques, des ensembles de données et des cadres de certification conformes aux normes européennes en matière de sécurité et de confidentialité, dans le but de réduire la dépendance vis-à-vis des fournisseurs d’IA non européens. Berlin indique que cette usine servira également de plateforme collaborative permettant aux instituts de recherche et à l’industrie de concevoir conjointement des systèmes d’IA sécurisés et interopérables pour les secteurs public et privé.

Pakistan. Le gouvernement pakistanais s’est engagé à réaliser d’importants investissements dans l’IA d’ici 2030, en déployant une stratégie nationale globale visant à accélérer la transformation numérique dans l’ensemble de l’économie. Ce plan vise à renforcer les capacités en matière d’IA dans des secteurs clés, notamment l’agriculture, la santé et l’éducation, grâce au financement de pôles de recherche, de partenariats public-privé et de programmes de perfectionnement ciblés. Selon les responsables, cet investissement a pour objectif d’attirer les investissements directs étrangers, de stimuler les exportations et de positionner le Pakistan comme un acteur technologique régional, tout en abordant les questions éthiques et les cadres de gouvernance afin d’orienter un déploiement responsable de l’IA.

Slovénie. La Slovénie a défini une vision nationale ambitieuse en matière d’IA, soulignant des priorités stratégiques telles qu’une IA centrée sur l’humain, des cadres éthiques solides et des investissements dans la recherche et les talents. La feuille de route met l’accent sur la collaboration avec les partenaires européens et le respect des normes internationales, positionnant la Slovénie comme un acteur proactif dans les discussions sur la gouvernance de l’IA.

PARTENARIATS

Corée du Sud et Singapour. La Corée du Sud et Singapour ont lancé une alliance Corée-Singapour pour l’IA, une initiative bilatérale visant à approfondir la coopération dans le domaine de l’IA et des technologies connexes. Annoncée lors du sommet Corée-Singapour AI Connect à Singapour, cette alliance vise à créer un écosystème d’innovation ouvert qui relie les capitaux, les talents et les technologies entre les deux nations, dans le but de renforcer la compétitivité sur le marché mondial de l’IA et de soutenir le développement conjoint de solutions d’IA qui répondent aux défis régionaux et mondiaux. Ce partenariat comprend l’engagement de créer d’ici 2030 un fonds d’investissement mondial de 300 millions de dollars américains à Singapour pour soutenir les start-ups et la recherche collaborative.

GOUVERNANCE DES CONTENUS

Chine. Un tribunal de l’est de la Chine a établi un précédent juridique en limitant la responsabilité des développeurs en cas d’hallucinations de l’IA, jugeant que les développeurs ne sont pas automatiquement responsables à moins que les utilisateurs ne puissent prouver une faute et un préjudice démontrable. Les juges ont qualifié les services d’IA de prestataires de services, exigeant des plaignants qu’ils démontrent à la fois la faute du prestataire et le préjudice réel causé par des résultats erronés, un cadre destiné à équilibrer les incitations à l’innovation et la protection des utilisateurs.

Les autorités chargées de la protection des données. Les autorités chargées de la protection des données de 61 juridictions et le Contrôleur européen de la protection des données (CEPD) ont publié une déclaration commune mettant en garde contre les outils d’IA qui génèrent des images réalistes de personnes identifiables sans leur consentement. Ils ont exprimé leurs préoccupations concernant la vie privée, la dignité et la sécurité des enfants, soulignant que ces technologies, souvent intégrées aux réseaux sociaux, permettent la diffusion d’images intimes non consenties et d’autres contenus préjudiciables. Les autorités ont souligné que les systèmes d’IA doivent se conformer aux lois sur la protection des données et que certaines utilisations peuvent constituer des infractions pénales. Les organisations ont été invitées à mettre en place des mesures de protection, à garantir la transparence, à permettre la suppression rapide des contenus et à collaborer de manière proactive avec les régulateurs afin de protéger les droits fondamentaux.

Inde. L’Inde a commencé à appliquer une règle de suppression dans les trois heures pour les contenus deepfake générés par l’IA, exigeant des plateformes et des intermédiaires qu’ils retirent les contenus spécifiés dans les 180 minutes suivant la notification, sous peine de sanctions réglementaires. Ce délai accéléré vise à freiner la propagation rapide de médias synthétiques trompeurs, dans un contexte de préoccupations accrues concernant la désinformation et les perturbations sociales.

Coalition mondiale pour la sécurité des enfants. Une large coalition de défenseurs des droits de l’enfant, d’organisations de sécurité numérique et de décideurs politiques a exhorté les gouvernements à interdire les outils d’IA de « nudification », demandant instamment la criminalisation des logiciels qui transforment des images de personnes habillées en versions sexuellement explicites sans leur consentement. Le groupe soutient que les approches existantes en matière de modération des contenus sont insuffisantes pour protéger les mineurs et souligne que des interdictions légales préventives sont nécessaires pour empêcher une exploitation généralisée.

UNICEF. Le Fonds des Nations Unies pour l’enfance (UNICEF) a exhorté les gouvernements à criminaliser la création, la possession et la distribution de contenus générés par l’IA représentant des abus sexuels sur des enfants, mettant en garde contre une forte augmentation des deepfakes sexuellement explicites impliquant des enfants et demandant instamment le renforcement des pratiques de sécurité dès la conception et une modération rigoureuse des contenus. Une étude citée par l’agence a révélé qu’au moins 1,2 million d’enfants dans 11 pays ont signalé que leurs images avaient été manipulées pour créer des deepfakes explicites à l’aide de l’IA, les outils de « nudification » qui déshabillent ou modifient les vêtements présentant des risques accrus. L’UNICEF a souligné que les deepfakes à caractère sexuel impliquant des mineurs devraient être traités comme du matériel pédopornographique au sens de la loi et a exhorté les plateformes numériques à empêcher leur diffusion plutôt que de se contenter de supprimer les contenus après coup.

Espagne. En Espagne, le Premier ministre Pedro Sánchez a ordonné aux procureurs d’enquêter sur X, Meta et TikTok pour la diffusion présumée de contenus pédopornographiques générés par l’IA (CSAM). L’enquête fait suite à des informations selon lesquelles les systèmes des plateformes auraient permis la création et la diffusion d’images deepfake à caractère sexuel impliquant des mineurs. Les autorités espagnoles examinent si les entreprises ont manqué à leur obligation d’empêcher la diffusion de tels contenus et si les outils d’IA intégrés ou liés aux plateformes ont contribué à ces préjudices.

Royaume-Uni. Le Royaume-Uni s’associe à Microsoft, à des universitaires et à des experts en technologie afin de développer un système de détection des deepfakes pour lutter contre les contenus préjudiciables générés par l’IA. Le cadre mis en place par le gouvernement permettra de normaliser la manière dont les outils de détection sont évalués par rapport aux menaces réelles telles que l’usurpation d’identité et l’exploitation sexuelle, en s’appuyant sur la législation récente qui criminalise la création d’images synthétiques intimes non consensuelles. Les responsables ont invoqué l’augmentation spectaculaire du nombre de deepfakes partagés en ligne ces dernières années pour justifier cette initiative.

Grok/X. La brigade chargée de la cybercriminalité du parquet de Paris a procédé à une perquisition dans les locaux français de X dans le cadre de cette enquête élargie. M. Musk et l’ancienne PDG Linda Yaccarino ont été convoqués pour des entretiens volontaires. X a nié toute malversation et a qualifié cette perquisition d’« acte abusif de la part des forces de l’ordre », tandis que M. Musk l’a décrite comme une « attaque politique ».

Le Bureau du commissaire à l’information du Royaume-Uni (ICO) a ouvert une enquête officielle sur X et xAI afin de déterminer si le traitement des données personnelles par Grok est conforme à la législation britannique en matière de protection des données, notamment aux principes fondamentaux de protection des données (légalité, équité et transparence), et si sa conception et son déploiement comprennent des protections intégrées suffisantes pour empêcher l’utilisation abusive des données personnelles à des fins de création d’images préjudiciables ou manipulées.

La Commission irlandaise de protection des données (DPC) a lancé une enquête à grande échelle au titre du RGPD sur le chatbot IA Grok de X, à la suite d’informations selon lesquelles ses capacités d’IA générative auraient été utilisées pour produire des contenus préjudiciables, non consensuels et à caractère sexuel impliquant des données à caractère personnel. Cette enquête, déclenchée par une controverse généralisée sur les images produites par Grok, se déroule parallèlement à la découverte de preuves indiquant que le chatbot a gagné des parts de marché aux États-Unis, alors que les régulateurs mondiaux examinent sa conformité aux normes fondamentales en matière de protection des données.

L’Agence nationale brésilienne de protection des données et le Bureau national des droits des consommateurs ont ordonné à X de cesser de fournir des images explicites via son IA Grok, invoquant les risques que des contenus préjudiciables atteignent des mineurs et la violation des normes locales en matière de sécurité numérique. La directive exige des mesures techniques immédiates pour bloquer certaines invites et certains contenus dans le cadre d’un examen continu des pratiques de modération du contenu de la plateforme.

Par ailleurs, l’Indonésie a rétabli l’accès à Grok après l’avoir interdit en janvier, ayant reçu de X l’assurance que des mesures de protection plus strictes seraient mises en place pour empêcher toute nouvelle utilisation abusive de l’outil d’IA.

Chili. Le Chili a lancé Latam-GPT afin de renforcer la présence de l’Amérique latine dans le domaine mondial de l’intelligence artificielle. Ce projet, développé par le Centre national pour l’intelligence artificielle avec le soutien de toute l’Amérique du Sud, vise à corriger les préjugés de longue date en formant les systèmes à partir des données propres à la région plutôt que de celles provenant principalement des États-Unis ou d’Europe. Le président Gabriel Boric a déclaré que ce modèle contribuera à préserver l’identité culturelle et permettra à la région de jouer un rôle plus actif dans le développement technologique. Latam-GPT n’est pas conçu comme un outil conversationnel, mais plutôt comme un vaste ensemble de données qui servira de base à de futures applications. Plus de huit téraoctets d’informations ont été collectés, principalement en espagnol et en portugais, et il est prévu d’ajouter des langues indigènes à mesure que le projet se développera.

SÛRETÉ ET SÉCURITÉ

Experts internationaux. Le deuxième rapport international sur la sécurité de l’IA 2026 a été publié. Ce rapport synthétise les données disponibles sur les capacités de l’IA, telles que l’amélioration du raisonnement et de la performance des tâches, ainsi que les risques émergents tels que les deepfakes, les cyber abus et la dépendance émotionnelle vis-à-vis des compagnons IA, tout en soulignant le manque de fiabilité et les défis persistants en matière de gestion des risques. Il vise à fournir aux décideurs politiques une base scientifique pour leurs décisions en matière de réglementation et de gouvernance, sans pour autant prescrire de politiques spécifiques.

L’ONU. La gouvernance de l’IA a été au cœur des discussions lors du récent dialogue spécial des Nations unies intitulé « Des principes à la pratique : dialogue spécial sur l’intelligence artificielle et la prévention et la lutte contre l’extrémisme violent ». Des diplomates et des experts ont discuté de la manière dont l’IA remodèle la stabilité mondiale, la dynamique des conflits et le droit international. Les participants ont souligné les risques liés aux systèmes autonomes et aux campagnes de désinformation, et ont insisté sur la nécessité d’une coopération multilatérale et de normes communes pour atténuer les menaces émergentes.

L’Europe. La Commission européenne a confirmé qu’elle reporterait à nouveau la publication des lignes directrices relatives aux systèmes d’IA à haut risque dans le cadre de la loi européenne sur l’IA. Ces lignes directrices devaient être publiées avant le 2 février 2026, mais elles suivront désormais un calendrier révisé. Ce report marque le deuxième délai non respecté et s’ajoute aux retards plus généraux pris dans la mise en œuvre de la loi européenne sur l’IA.

DROITS DE PROPRIÉTÉ INTELLECTUELLE

Le Royaume-Uni. La Cour suprême britannique a statué que les inventions assistées par l’IA peuvent être brevetées lorsque le rôle inventif de l’humain est identifiable et substantiel. Selon les experts juridiques, cette décision favorisera l’innovation en clarifiant la protection de la propriété intellectuelle dans le domaine du développement hybride humain-IA. Le jugement vise à encourager les investissements dans la recherche sur l’IA tout en maintenant les normes de brevetabilité établies.

L’AVENIR DU TRAVAIL

Corée du Sud. La Corée du Sud a mis en place un organisme gouvernemental chargé de traiter les pressions exercées par l’automatisation liée à l’intelligence artificielle sur la main-d’œuvre, créant ainsi un conseil intersectoriel chargé de prévoir les tendances en matière de suppression d’emplois et de recommander des mesures politiques. Cette initiative rassemble des syndicats, des dirigeants industriels et des ministères afin de coordonner les programmes de reconversion et de perfectionnement professionnels, de renforcer les dispositifs de protection sociale et d’étudier des modèles de soutien au revenu pour les travailleurs touchés par l’automatisation.

Des procès phares débutent alors que de plus en plus de pays envisagent d’interdire l’accès aux réseaux sociaux aux enfants

Les procès commencent

Le procès de Los Angeles sur l’addiction aux réseaux sociaux. Un procès historique s’est ouvert à Los Angeles, aux États-Unis, en février 2026 contre Meta et YouTube, centré sur des allégations selon lesquelles leurs plateformes sont délibérément conçues pour créer une dépendance et ont nui à la santé mentale des jeunes utilisateurs.

La plaignante, Kaley, âgée de 20 ans en 2026, affirme qu’Instagram et YouTube lui ont causé de l’anxiété, une dysmorphie corporelle et des pensées suicidaires.

Ses avocats ont comparé les fonctionnalités telles que le défilement infini, la lecture automatique, les likes et les filtres beauté à un « casino numérique » pour enfants, citant des documents internes montrant que les plateformes ciblaient les jeunes utilisateurs et utilisaient même YouTube comme une « baby-sitter numérique ». Kaley avait initialement également poursuivi Snap Inc. et TikTok, mais ces plateformes ont conclu des accords confidentiels avant le procès, laissant Meta et YouTube comme seuls défendeurs à comparaître devant un jury.

La défense de Meta et YouTube a fait valoir que les réseaux sociaux n’étaient pas responsables des difficultés de Kaley, citant son milieu familial difficile, les dossiers des thérapeutes et la disponibilité d’outils de sécurité.

YouTube a souligné que l’utilisation quotidienne moyenne de Kaley était de 29 minutes depuis 2020 et a comparé la plateforme à d’autres services de divertissement, soulignant qu’elle n’était pas dépendante.

Le PDG de Meta, Mark Zuckerberg, a témoigné et a insisté sur le fait qu’Instagram interdit les utilisateurs de moins de 13 ans et qu’il est difficile de faire respecter les limites d’âge, car de nombreux mineurs mentent sur leur date de naissance. Il a souligné les efforts continus visant à réduire le temps passé devant les écrans et à améliorer les fonctionnalités de sécurité. Néanmoins, des documents internes présentés au tribunal suggèrent que l’engagement des jeunes adolescents était une priorité stratégique. Cette affaire est suivie de près, car elle pourrait servir de modèle pour la responsabilité des plateformes en matière de fonctionnalités addictives.

La psychiatre de la plaignante, Virginia Burke, a également témoigné. Mme Burke a déclaré que l’utilisation des réseaux sociaux par la plaignante Kaley avait contribué à ses problèmes de santé mentale, citant notamment le harcèlement en ligne. Cependant, Mme Burke a noté que Kaley aimait également créer et partager des vidéos artistiques, même si elle était frustrée lorsque d’autres s’en attribuaient le mérite. Mme Burke a déclaré que la dépendance aux réseaux sociaux n’était pas encore un diagnostic largement reconnu en psychiatrie et qu’elle ne figurait pas dans le dernier Manuel diagnostique et statistique, le texte de référence pour les professionnels de la santé mentale aux États-Unis.

Kaley elle-même a témoigné que dès son plus jeune âge, elle passait presque tout son temps sur des plateformes telles que YouTube et Instagram, décrivant son incapacité à limiter son utilisation, même lorsqu’elle était victime de harcèlement. Elle a déclaré qu’elle récupérait secrètement son téléphone la nuit après que sa mère le lui ait confisqué, qu’elle était bouleversée lorsqu’on lui refusait l’accès aux applications de réseaux sociaux, qu’elle se retirait des interactions familiales et qu’elle pensait que sa santé, son sommeil, ses notes et son bien-être général auraient été meilleurs sans les réseaux sociaux.

Le procès devrait durer jusqu’à la fin mars 2026.

Le procès pour la protection des consommateurs à Santa Fe. Un autre procès s’est ouvert à Santa Fe, au Nouveau-Mexique, aux États-Unis, après plus de deux ans de débats préliminaires. Le procès, intenté en 2023 par le procureur général du Nouveau-Mexique, Raúl Torrez, a été porté devant les tribunaux et un jury.

La plainte accuse Meta d’avoir enfreint les lois de l’État sur la protection des consommateurs en présentant de manière trompeuse la sécurité de ses plateformes pour les mineurs, tout en développant des fonctionnalités et des algorithmes qui, selon les procureurs, incitent à une utilisation prolongée et exposent les enfants à des risques importants. Ces risques comprennent une dépendance, l’exposition à des contenus sexuels préjudiciables, des communications privées non désirées avec des adultes, des troubles du sommeil dus à une utilisation compulsive et des environnements où les prédateurs peuvent agir relativement facilement.

Le 3 mars, les procureurs de l’État ont présenté les dépositions enregistrées du PDG de Meta, Mark Zuckerberg, et du directeur d’Instagram, Adam Mosseri, cherchant à démontrer que l’entreprise était consciente des mesures et des recherches indiquant de graves problèmes de sécurité pour les enfants, mais qu’elle n’avait pas suffisamment agi ni averti les utilisateurs et les familles. Dans ces dépositions, les procureurs ont interrogé les dirigeants de Meta sur des questions telles que les priorités en matière de sécurité par rapport aux profits de l’entreprise, la portée de leurs plateformes auprès des adolescents et les choix spécifiques de produits — des systèmes de recommandation aux filtres cosmétiques — susceptibles d’affecter le bien-être des adolescents.

Les avocats de Meta ont déclaré au jury lors de leurs déclarations liminaires que l’entreprise avait mis en place de nombreux outils de sécurité et systèmes de modération des contenus. Meta soutient qu’elle n’a pas commis de tromperie et qu’elle a toujours divulgué les risques et pris des mesures de sécurité.

Le procès des districts scolaires d’Oakland. Un autre procès phare devrait débuter en juin à Oakland, en Californie. Il s’agit du premier procès intenté par des districts scolaires contre des plateformes de réseaux sociaux pour préjudice causé à des enfants.

Le fond du problème. Environ 1 600 plaignants, dont des particuliers, plus de 350 familles, 250 districts scolaires et des procureurs généraux d’État, ont intenté des poursuites contre Meta Platforms, Snap Inc., TikTok et Google. En raison de leur nombre, ces poursuites sont coordonnées dans le cadre d’une procédure de coordination du Conseil judiciaire (JCCP). Au sein de ce groupe coordonné, 22 procès phares ont été sélectionnés, dont trois — à Los Angeles, Santa Fe et Oakland — devraient être jugés en premier.

Un procès phare est un procès test choisi parmi un grand nombre de procès similaires pour être jugé en premier, afin d’évaluer la réaction des jurys face aux preuves et aux arguments juridiques. Son verdict ne tranche pas les autres affaires, mais il donne une indication sur la manière dont les futurs procès ou négociations de règlement pourraient se dérouler.

C’est pourquoi les procès de Los Angeles et de Santa Fe ont suscité beaucoup d’attention : leur verdict devrait influencer les futures pratiques en matière de conception des plateformes.

Jury, Person, People, Crowd, Face, Head, Audience

Le nombre de pays interdisant l’accès aux réseaux sociaux aux mineurs augmente

La tendance visant à interdire aux enfants l’accès aux réseaux sociaux se poursuit, dix autres pays envisageant des mesures législatives et des outils de mise en application.

Le parlement portugais a approuvé une loi restreignant l’accès aux réseaux sociaux pour les mineurs de moins de 16 ans, exigeant le consentement explicite et vérifié des parents pour accéder à des plateformes telles qu’Instagram, TikTok et Facebook. L’accès sera contrôlé par le biais de la clé mobile numérique, le système national d’identification numérique du Portugal, garantissant une vérification efficace de l’âge et la conformité des plateformes. Cette loi renforce les protections face aux préoccupations croissantes concernant l’impact des réseaux sociaux sur la santé mentale des jeunes. Les règles détaillées de mise en œuvre et d’application sont désormais soumises à l’examen de la commission parlementaire.

En Espagne, le gouvernement du Premier ministre Pedro Sánchez a proposé une législation qui interdirait l’accès aux réseaux sociaux aux utilisateurs de moins de 16 ans, présentant cette mesure comme un outil nécessaire à la protection des enfants contre la dépendance, l’exploitation et les contenus préjudiciables. Selon le projet de plan, les plateformes doivent mettre en place des systèmes obligatoires de vérification de l’âge conçus comme des barrières applicables plutôt que comme des mesures de protection symboliques, ce qui marque un changement vers une application plus stricte de la réglementation plutôt que vers une conformité volontaire de la part des entreprises technologiques. Les propositions incluent également la responsabilité juridique des dirigeants du secteur technologique en cas de contenu illégal ou haineux restant en ligne.

Le parti AK au pouvoir en Turquie a proposé un projet de loi interdisant l’accès aux réseaux sociaux aux enfants de moins de 15 ans, obligeant les plateformes à mettre en place une vérification de l’âge. Les plateformes devront également créer une version distincte de la plateforme spécialement destinée aux mineurs âgés de 15 à 18 ans. Les responsables ont invoqué le code pénal turc, qui limite la responsabilité pénale des enfants de moins de 15 ans, pour justifier la limite d’âge fixée à 15 ans.

La coalition au pouvoir en Pologne est actuellement en train d’élaborer une loi qui interdirait l’utilisation des réseaux sociaux aux enfants de moins de 15 ans. Les législateurs souhaitent finaliser la loi d’ici la fin février 2026 et la mettre en œuvre potentiellement d’ici Noël 2027. La Pologne souhaite mettre à jour son application d’identité numérique, mObywatel, afin de permettre aux utilisateurs de vérifier leur âge.

La Slovénie prépare actuellement un projet de loi visant à interdire aux mineurs de moins de 15 ans d’accéder aux réseaux sociaux, une initiative lancée par le ministère de l’Éducation.

La Grèce serait sur le point d’annoncer une interdiction d’utilisation des réseaux sociaux pour les enfants de moins de 15 ans. Le ministère de la Gouvernance numérique a l’intention de s’appuyer sur l’application Kids Wallet, lancée l’année dernière, comme mécanisme de mise en œuvre de la mesure, plutôt que de développer un nouveau cadre de contrôle.

En Autriche, le gouvernement débat activement d’une interdiction d’utilisation des réseaux sociaux pour les enfants de moins de 14 ans. Le secrétaire d’État aux Affaires numériques, Alexander Pröll, a confirmé que cette politique était en cours de discussion dans le but de la mettre en œuvre d’ici la rentrée scolaire de septembre 2026.

L’Allemagne envisage de limiter l’accès des enfants aux réseaux sociaux, le parti au pouvoir exhortant le gouvernement fédéral à introduire un âge minimum légal de 14 ans. Le chancelier Friedrich Merz a signalé son soutien à cette proposition, se déclarant très favorable à cette idée.

Le Royaume-Uni a lancé une consultation publique afin de déterminer si les enfants de moins de 16 ans devraient faire l’objet de restrictions ou d’une éventuelle interdiction d’utiliser les réseaux sociaux. Les jeunes, les parents et les éducateurs sont invités à partager leurs points de vue avant que les ministres ne se prononcent sur la politique future. Les mesures envisagées pourraient inclure la fixation d’un âge minimum pour l’utilisation des réseaux sociaux, la restriction des fonctionnalités préjudiciables telles que le défilement infini et l’examen des protections contre l’envoi ou la réception d’images explicites par les enfants. La consultation explorera également les restrictions à l’utilisation des chatbots IA par les enfants et les limites à l’utilisation des VPN lorsqu’ils compromettent les mesures de sécurité. Le gouvernement a l’intention d’agir rapidement sur la base de ses conclusions dans les mois à venir en introduisant des pouvoirs juridiques ciblés qui pourront être mis en œuvre rapidement à mesure que la technologie évolue.

L’UE dans son ensemble réexamine l’idée d’une restriction d’âge pour les réseaux sociaux à l’échelle européenne. La question a été soulevée dans le nouveau plan d’action de la Commission européenne contre le cyberharcèlement, publié le mardi 10 février. Ce plan confirme qu’un groupe d’experts en protection de l’enfance conseillera la Commission d’ici l’été sur d’éventuelles restrictions d’âge à l’échelle de l’UE pour l’utilisation des réseaux sociaux. Le groupe évaluera les options pour une approche européenne coordonnée, y compris une éventuelle législation et des mesures de sensibilisation des parents.

Le document note que des règles nationales divergentes pourraient conduire à une protection inégale des enfants dans l’ensemble de l’Union. Selon la Commission, un cadre européen harmonisé contribuerait à garantir des garanties cohérentes et à réduire la fragmentation dans la manière dont les plateformes appliquent les restrictions d’âge.

Ces efforts nationaux individuels s’inscrivent dans un contexte de coordination réglementaire internationale croissante. Le 3 février 2026, la Commission européenne s’est réunie avec le commissaire australien à la sécurité électronique et l’Ofcom britannique afin de partager leurs points de vue sur les mesures de vérification de l’âge, c’est-à-dire les approches techniques et politiques permettant de vérifier l’âge des utilisateurs et d’appliquer des restrictions en ligne adaptées à leur âge.

Cette réunion faisait suite à une communication conjointe signée à la fin de 2025, dans laquelle les trois régulateurs s’engageaient à poursuivre leur collaboration afin de renforcer la sécurité en ligne des enfants, notamment en explorant les technologies efficaces de vérification de l’âge, les stratégies d’application et le rôle des données et de la recherche indépendante dans l’action réglementaire.

Vue d’ensemble. Le nombre de membres du club des interdictions a atteint les deux chiffres. Nous continuerons à suivre l’évolution de la situation.

Zooming arrière. Ces initiatives menées dans plusieurs pays confirment que l’interdiction des réseaux sociaux en Australie n’était pas une expérience politique isolée, mais plutôt le début d’un effet d’entraînement mondial. Cette dynamique est particulièrement frappante étant donné que l’interdiction australienne n’est pas encore largement considérée comme un succès : son efficacité et ses répercussions plus larges font encore l’objet d’études et de débats.

Ces développements interviennent alors que le rapport eSafety australien souligne que les géants de la technologie, notamment Apple, Google, Meta, Microsoft, Discord, Snap, Skype et WhatsApp, n’ont réalisé que des progrès limités dans la lutte contre l’exploitation et les abus sexuels des enfants en ligne (CSEA), bien qu’ils soient légalement tenus de rendre compte des mesures prises en vertu de la loi australienne sur la sécurité en ligne.

Des interdictions à la conformité

Au-delà des interdictions pures et simples, un deuxième front réglementaire se dessine : le renforcement des conditions juridiques et techniques dans lesquelles les plateformes traitent les données des mineurs et hébergent des contenus susceptibles de nuire aux enfants.

Parallèlement, les mesures coercitives prises au Royaume-Uni soulignent les risques financiers et les atteintes à la réputation liés au non-respect de la réglementation. L’autorité britannique de protection de la vie privée a infligé une amende de 20 millions de livres sterling à Reddit pour avoir traité illégalement les données personnelles d’enfants et ne pas avoir protégé les utilisateurs de moins de 13 ans. L’autorité de régulation a estimé que Reddit ne disposait pas de « mécanismes fiables de vérification de l’âge » et s’appuyait sur une auto-déclaration facilement contournable, ce qui signifie qu’il n’avait aucune base légale pour traiter les données des enfants et les exposait à des contenus potentiellement préjudiciables. Reddit n’a pas non plus réalisé l’évaluation d’impact sur la protection des données requise avant 2025. Cette amende est la plus importante infligée par l’ICO en matière de protection de la vie privée des enfants. Reddit prévoit de faire appel.

L’autorité chargée de la protection des données en Turquie a ouvert une nouvelle enquête sur la manière dont les principales plateformes de réseaux sociaux gèrent les données personnelles des enfants. L’Autorité de protection des données personnelles examine actuellement la manière dont les données personnelles des enfants sont traitées sur TikTok, Instagram, Facebook, YouTube, X et Discord, ainsi que les mesures de protection mises en place. Par ailleurs, le Parti de la justice et du développement (AKP), au pouvoir, devrait introduire un dispositif familial qui exigerait la vérification de l’identité de chaque compte par le biais de numéros de téléphone ou du système e-Devlet. Les enfants de moins de 15 ans ne seraient pas autorisés à créer de profil, et d’autres restrictions pourraient s’appliquer aux utilisateurs de moins de 18 ans.

Au Brésil, une nouvelle proposition législative (projet de loi n° 730/2026) a été présentée à la Chambre des députés. Cette proposition impose des mécanismes de vérification de l’âge conformes à la loi brésilienne sur la protection des données (loi n° 13.709/2018), qui donne la priorité à la minimisation des données, à la pseudonymisation et à la conservation limitée. Elle interdit également la monétisation directe ou indirecte impliquant des enfants de moins de 14 ans et exige une autorisation judiciaire préalable pour les travaux artistiques rémunérés des adolescents âgés de 14 à 18 ans, sous réserve de garanties. S’il est adopté, le projet de loi officialiserait les exigences de conformité de base pour les plateformes opérant sur le marché brésilien.

Les mesures provisoires chinoises qui classifient et réglementent les informations en ligne susceptibles de nuire à la santé physique et mentale des mineurs sont entrées en vigueur le 1er mars. Les contenus concernés comprennent les contenus incitant à des comportements dangereux, à des émotions extrêmes, à la discrimination, à des modes de vie malsains, à une consommation irrationnelle, au culte des célébrités ou à des valeurs déformées telles que l’hédonisme et la pseudoscience. Les règles limitent également l’utilisation abusive des images et des données personnelles des mineurs. Les fournisseurs de systèmes de recommandation algorithmiques et de services d’IA générative sont tenus de renforcer et d’affiner leurs cadres de gouvernance en matière de sécurité et leurs mesures de protection techniques, et il leur est interdit de promouvoir ou de diffuser des contenus en ligne susceptibles d’avoir un impact négatif sur le bien-être physique ou psychologique des mineurs.

Pourquoi le cyberespace n’existe pas

Il y a trente ans, le 8 février 1996, deux événements ont donné lieu à un discours influent sur Internet, affirmant qu’il occupait un domaine distinct du droit et de la politique ordinaires. Il s’agit de la Déclaration d’indépendance du cyberespace et de la loi américaine sur la décence dans les communications (Communications Decency Act, CDA).

Déclaration d’indépendance du cyberespace. À Davos, la Déclaration d’indépendance du cyberespace de John Perry Barlow affirmait que les « gouvernements du monde industriel » n’avaient « aucune souveraineté » dans le cyberespace.

Cette vision a donné naissance à une génération de penseurs qui affirmaient qu’Internet signifiait « la fin de la géographie ». Des milliers d’articles, de livres, de thèses et de discours ont été publiés pour soutenir que nous avons besoin d’une nouvelle gouvernance pour le « nouveau monde courageux » du numérique.

Ce château de cartes intellectuel et politique reposait sur l’hypothèse qu’il existe un cyberespace au-delà de l’espace physique. C’était (et c’est toujours) une hypothèse erronée. Le cyberespace n’existe pas. Chaque e-mail, chaque publication, chaque requête d’IA est en fin de compte un événement physique : des impulsions d’électrons transportant des bits et des octets à travers des câbles sous-marins, le Wi-Fi, des serveurs de données et l’infrastructure Internet.

The CDA and its Section 230. Le jour même de la déclaration de Barlow, le président Clinton a promulgué la loi américaine sur la décence dans les communications (US Communications Decency Act, CDA), qui avait été adoptée par le Congrès américain. Cette loi comprenait la section 230, qui accordait aux plateformes Internet une immunité sans précédent : elles ne pouvaient être considérées comme des éditeurs ou des diffuseurs du contenu qu’elles hébergeaient.

Pour la première fois dans l’histoire, les entités commerciales se voyaient accorder une large protection contre toute responsabilité pour l’activité même dont elles tiraient profit. Il s’agissait d’une rupture avec la longue tradition de responsabilité juridique, par exemple celle d’un journal pour les textes qu’il publie ou celle des diffuseurs pour leurs émissions.

Cette disposition était justifiée comme un moyen de protéger une industrie naissante contre des litiges paralysants. À l’époque, les entreprises Internet étaient petites et expérimentales. L’immunité a permis une croissance et une innovation rapides.

Au fil du temps, cependant, ces start-ups sont devenues certaines des entreprises les plus valorisées de l’histoire, avec une portée mondiale et des capitalisations boursières de plusieurs milliers de milliards de dollars. Le cadre juridique est toutefois resté largement inchangé, alors même que les entreprises Internet ont développé des algorithmes sophistiqués qui sélectionnent, amplifient et monétisent le contenu des utilisateurs à grande échelle. Cette divergence a créé une tension centrale dans le droit et l’économie contemporains : des intermédiaires extrêmement puissants opérant avec une responsabilité limitée quant aux effets systémiques.

La convergence des deux. La séparation conceptuelle du « cyberespace » a facilité la défense de cet arrangement. Si Internet était un nouveau monde, des règles exceptionnelles semblaient justifiées.

Cependant, les critiques ont rapidement contesté ce raisonnement. Le juge américain Frank H. Easterbrook a fait valoir que nous n’avions pas besoin d’une loi sur Internet, tout comme nous n’avions pas besoin d’une « loi sur les chevaux » lorsque ceux-ci ont été introduits comme mode de transport dominant. Internet devrait être réglementé en appliquant les principes juridiques existants. Le droit régit les relations entre les personnes et les institutions, quelles que soient les technologies qu’elles utilisent. Le support peut changer, mais les principes sous-jacents demeurent.

L’expérience a largement confirmé ce point de vue. Les technologies numériques n’ont pas effacé les frontières géographiques, elles les ont renforcées. Les États affirment leur juridiction sur les flux de données, la modération des contenus, la fiscalité, la concurrence et la sécurité. La géolocalisation de haute précision, les exigences en matière de localisation des données et les régimes réglementaires nationaux démontrent qu’Internet fonctionne strictement dans les limites territoriales.

Cependant, la CDA reste en vigueur et s’étend à l’ère de l’IA. Les entreprises qui développent de grands modèles linguistiques et d’autres systèmes d’IA s’appuient souvent sur des protections intermédiaires et des doctrines analogues pour limiter leur responsabilité. En conséquence, les outils d’IA peuvent être déployés à l’échelle mondiale avec une surveillance ex ante relativement limitée. Pourtant, leurs résultats peuvent façonner le discours public, influencer les élections, affecter la santé mentale et générer des perturbations économiques.

La question centrale n’est pas de savoir si l’innovation doit être restreinte, mais si elle doit être alignée sur les principes établis de responsabilité. Les technologies n’existent pas en dehors de la société ; elles y sont intégrées. Si une entité conçoit, déploie et tire profit d’un système, elle doit assumer la responsabilité de ses impacts prévisibles. L’ère de l’exceptionnalisme juridique doit prendre fin.

Les gouvernements poursuivent leurs efforts en faveur de la souveraineté numérique

Le mois dernier a été marqué par de nouveaux développements indiquant que la souveraineté numérique est la tendance dominante, qui s’est poursuivie de décembre 2025 à janvier et février 2026.

La Commission européenne a commencé à tester le protocole open source Matrix comme alternative possible aux plateformes de messagerie propriétaires pour la communication interne. L’architecture fédérée de Matrix permet d’héberger les communications sur des infrastructures européennes et de les régir selon les règles de l’UE, ce qui s’inscrit dans le cadre d’efforts plus larges visant à mettre en place des services publics numériques souverains et à réduire la dépendance à l’égard des plateformes externes.

La Commission a également dévoilé EURO-3C, une nouvelle initiative d’une valeur de 75 millions d’euros dans le cadre du programme Horizon Europe visant à construire la première infrastructure européenne à grande échelle de télécommunications fédérées et de cloud de pointe. En fédérant les infrastructures nationales existantes au-delà des frontières, EURO-3C vise à réduire la dépendance vis-à-vis des géants mondiaux hors UE et à renforcer le rôle de l’UE dans le cloud, l’edge computing et les infrastructures d’IA.

En France, le gouvernement a adopté une position ferme en matière de contrôle des infrastructures satellitaires, autre pilier de la souveraineté numérique. Paris a bloqué la vente des actifs des stations terrestres appartenant à Eutelsat à un investisseur externe, arguant que ces infrastructures soutiennent les communications spatiales civiles et militaires et doivent rester sous autorité nationale. Les responsables français ont décrit ces installations comme essentielles à l’autonomie stratégique, en partie parce qu’Eutelsat représente l’un des rares concurrents européens des constellations de satellites américaines telles que Starlink.

En Russie, l’autorité de régulation des télécommunications Roskomnadzor a renforcé les restrictions imposées à Telegram, ralentissant la diffusion des médias et limitant certaines fonctionnalités afin d’inciter les utilisateurs à se tourner vers des alternatives nationales. Roskomnadzor a déclaré que Telegram ne prenait pas de mesures significatives pour lutter contre la fraude, ne protégeait pas les données personnelles des utilisateurs et enfreignait les lois russes. Le fondateur de Telegram a critiqué ces mesures, les qualifiant d’autoritaires et avertissant qu’elles pourraient nuire aux services de communication essentiels.

Cette répression s’est intensifiée avec le blocage complet de WhatsApp, propriété de Meta, utilisé par 100 millions de Russes. Les autorités ont justifié cette interdiction en invoquant le refus de WhatsApp de se conformer aux exigences légales russes. Les utilisateurs sont encouragés à adopter des plateformes soutenues par le gouvernement qui, selon les critiques, permettent la surveillance de l’État, ce qui soulève des inquiétudes quant à la confidentialité et à l’accès à des canaux de communication indépendants. Meta a qualifié cette interdiction de préjudiciable à la sécurité et à la confidentialité.

Malgré ces mesures, la Russie a décidé de suspendre ses actions agressives contre Google, invoquant la dépendance du pays à l’égard des appareils Android et avertissant qu’une interdiction soudaine pourrait perturber des millions d’utilisateurs. Les responsables ont indiqué que toute transition vers des alternatives nationales se ferait progressivement, reflétant une approche prudente visant à réduire la dépendance à l’égard des technologies étrangères.

Pendant ce temps, aux Pays-Bas, la souveraineté numérique est devenue un sujet central des débats parlementaires. Les législateurs ont renouvelé leurs appels en faveur du transfert des données des secteurs public et privé hors des services cloud basés aux États-Unis, invoquant les risques liés à la législation américaine telle que le Cloud Act. Les inquiétudes se sont intensifiées à la suite du projet d’acquisition de Solvinity, qui héberge une partie du système d’identité numérique néerlandais DigiD, par une entreprise américaine. Les députés ont souligné la nécessité de renforcer les garanties, de promouvoir des alternatives européennes ou néerlandaises au cloud et de mettre à jour les règles de passation des marchés publics afin de protéger les données sensibles.

Alors que les décideurs politiques européens évaluent l’autonomie stratégique et le contrôle réglementaire, Washington intensifie simultanément ses efforts pour contrer ce qu’il considère comme des mesures potentiellement perturbatrices au niveau mondial.

Un câble interne du département d’État américain consulté par Reuters demande aux diplomates américains de s’opposer activement aux lois étrangères sur la souveraineté et la localisation des données et de promouvoir une politique internationale plus affirmée des États-Unis en matière de données. Ce câble, signé le 18 février, fait valoir que de telles réglementations pourraient perturber les flux de données mondiaux, augmenter les coûts, créer des exigences de conformité inutilement lourdes et entraver les services de cloud computing et d’intelligence artificielle. La directive encourage également la promotion du Global Cross-Border Privacy Rules Forum (CBPR) comme mécanisme alternatif soutenant les flux de données avec des protections de la vie privée.

Zoom arrière. Cette décision souligne les tensions croissantes avec l’Europe concernant la réglementation en matière de confidentialité et de souveraineté numérique, et reflète une volonté de défendre les intérêts technologiques américains à l’étranger.

Vue d’ensemble. Le fil conducteur est clair : la souveraineté numérique est désormais une considération essentielle pour les gouvernements du monde entier. Les approches peuvent différer, mais l’objectif reste largement le même : garantir que l’avenir numérique d’une nation soit façonné par ses propres priorités et règles. Cependant, la véritable indépendance est entravée par des chaînes d’approvisionnement mondiales profondément ancrées, des coûts prohibitifs pour la mise en place de systèmes parallèles et le risque d’étouffer l’innovation par l’isolement. Si la volonté stratégique de souveraineté est claire, se détacher des écosystèmes technologiques interdépendants nécessitera des années d’investissement, de migration et d’adaptation. Les initiatives actuelles marquent le début d’une transition longue et difficile.

Quand l’éthique de l’IA entre en conflit avec la sécurité nationale : le différend entre Anthropic et le Pentagone

Dans un affrontement à haut risque qui redessine les lignes de front entre la Silicon Valley et l’armée américaine, la société d’intelligence artificielle Anthropic s’est retrouvée exclue du Pentagone après avoir refusé de renoncer à ses garanties éthiques.

Comment cela a commencé : un différend contractuel. Le Pentagone a demandé à Anthropic de garantir que son modèle, Claude, pourrait être utilisé à « toutes fins légales ». Anthropic a refusé, arguant que cette formulation ne limitait pas suffisamment les utilisations que l’entreprise considère comme à haut risque, en particulier la surveillance domestique de masse et les systèmes d’armes entièrement autonomes. L’entreprise a demandé des garde-fous plus clairs.

Comment la situation s’est aggravée : une menace pour la chaîne d’approvisionnement. Les responsables ont indiqué qu’Anthropic risquait de perdre ses contrats fédéraux et pourrait même être désignée comme un risque pour la chaîne d’approvisionnement. L’administration a finalement donné suite à cette mesure, ordonnant aux agences de cesser d’utiliser les systèmes d’Anthropic et accordant une période de transition limitée à six mois pour les accords existants.

Les désignations de risque pour la chaîne d’approvisionnement concernent généralement les menaces étrangères, et non les entreprises nationales qui négocient les termes d’un contrat. Les experts et Anthropic affirment qu’une telle désignation n’a pas de précédent clair et qu’elle est susceptible d’être contestée devant les tribunaux pour cause d’illégalité.

Entrée en scène : la société concurrente.C’est dans ce vide qu’OpenAI est intervenu. Peu après l’inscription d’Anthropic sur la liste noire, OpenAI a conclu son propre accord avec le Pentagone. La société a publiquement souligné qu’elle maintenait des « lignes rouges » en matière de sécurité, notamment des restrictions liées à la surveillance de masse et l’exigence d’un contrôle humain dans l’usage de la force. Le PDG Sam Altman a indiqué qu’OpenAI partageait bon nombre des préoccupations éthiques soulevées par Anthropic. Le Pentagone a accepté le cadre proposé par OpenAI, ce qui soulève des questions quant à la raison pour laquelle des garanties similaires se sont avérées intenables dans le cas d’Anthropic.

Les réactions. Cet épisode a eu des répercussions dans tout le monde technologique. De grands groupes industriels, dont des représentants d’Amazon, de Nvidia, d’Apple et d’autres, ont mis en garde le gouvernement contre une utilisation généralisée des désignations de risque pour la chaîne d’approvisionnement des entreprises technologiques américaines, craignant des effets dissuasifs sur l’innovation et la coopération entre les secteurs public et privé.

Les investisseurs d’Anthropic ont également fait pression pour apaiser les tensions, craignant que l’impasse ne nuise aux activités commerciales de l’entreprise et à ses perspectives d’introduction en bourse si des contrats clés étaient perdus.

Cependant, le lendemain de la perte du contrat avec le Pentagone par Anthropic, Claude s’est hissé à la première place des téléchargements d’applications aux États-Unis, tandis que les désinstallations de l’application mobile ChatGPT ont bondi de 295 % d’un jour à l’autre. Cela suggère que Claude pourrait ne pas perdre en popularité auprès des utilisateurs.

Pourquoi est-ce important ? En fin de compte, ce différend a mis en évidence une tension structurelle plus profonde. Les systèmes d’IA avancés occupent une place de plus en plus centrale dans la planification militaire, la logistique, l’analyse du renseignement et la prise de décision sur le champ de bataille. Dans le même temps, les principales entreprises d’IA ont défini des limites éthiques en matière de surveillance, d’autonomie létale et de risques liés au double usage.

La confrontation entre Anthropic et le Pentagone a cristallisé la question de savoir qui détermine ces limites lorsque la sécurité nationale et la gouvernance d’entreprise entrent en conflit.

Cependant, alors qu’Anthropic et le Pentagone seraient de retour à la table des négociations, cela souligne que la question est loin d’être résolue.tection mises en œuvre jusqu’à présent, soulignant que les solutions pourraient devoir être adaptées aux différentes juridictions.

Le mois dernier à Genève

Comprendre le groupe de travail d’UNCTAD sur la gouvernance des données | Partie 3

Le 10 février, Diplo, l’Open Knowledge Foundation et la Geneva Internet Platform ont coorganisé un événement en ligne intitulé « Décrypter le groupe de travail d’UNCTAD sur la gouvernance des données | Partie 3 », qui a examiné les progrès et les perspectives du groupe de travail multipartite des Nations unies sur la gouvernance des données. Les participants ont indiqué que les discussions s’étaient intensifiées ces derniers mois après des retards procéduraux initiaux. Cependant, de profondes divergences subsistent. L’une des lignes de fracture concerne la question de savoir si la gouvernance des données doit se concentrer principalement sur la protection de la vie privée des individus ou intégrer des droits sociétaux et collectifs plus larges en matière de données. Une autre porte sur la question de savoir si l’interopérabilité doit être considérée comme un bien public intrinsèque ou évaluée à la lumière des risques potentiels tels que la concentration du marché et l’extractivisme des données. Un troisième sujet de controverse concerne les flux transfrontaliers de données. Certains considèrent que la libre circulation des données dans un climat de confiance est essentielle à l’intégration dans les marchés mondiaux de l’innovation.

D’autres soutiennent que les asymétries en matière d’infrastructures et de pouvoir de négociation nécessitent de préserver l’autonomie réglementaire des pays en développement, y compris la capacité de mener des politiques de souveraineté des données alignées sur les priorités nationales de développement. Les intervenants ont souligné que le renforcement des capacités est devenu un domaine de convergence croissante. Les gouvernements de toutes les régions ont appelé à un soutien technique, institutionnel et politique pour développer des systèmes interopérables et renforcer les cadres nationaux de gouvernance des données. Bien que le rapport final sera succinct, les membres ont indiqué qu’il refléterait les divergences de vues sans privilégier les positions majoritaires.

Les discussions se poursuivent pour déterminer si le résultat final doit inclure des recommandations ou rester descriptif.

Le groupe de travail doit se réunir à nouveau à Genève en mars, alors qu’il s’apprête à finaliser un document qui, selon ses promoteurs, devrait guider les débats internationaux sur la gouvernance des données dans les années à venir.

Menaces hybrides: Comment améliorer la résilience face à ce phénomène?

L’Institut des Nations Unies pour la recherche sur le désarmement (UNIDIR), en partenariat avec l’Organisation internationale de la Francophonie (OIF), a organisé un événement afin d’examiner le phénomène des menaces hybrides. Les experts ont recommandé de renforcer la gouvernance multilatérale grâce à des normes harmonisées pour l’espace et les plateformes en ligne, de renforcer la résilience sociétale par l’information et l’application de la loi, et de protéger les infrastructures critiques par la cybersécurité et des mesures de protection opérationnelles. Il est essentiel de soutenir les États moins bien équipés en leur fournissant des outils techniques, réglementaires et de gestion des risques, tout en envoyant des signaux stratégiques pour clairement montrer les conséquences d’actions inacceptables. Le principe directeur de toutes ces mesures est l’intégration : les domaines spatial, informationnel et cybernétique doivent être gérés conjointement afin de maintenir la stabilité et la résilience mondiales.

Lancement du Rapport mondial sur la propriété intellectuelle 2026 : La technologie en mouvement

L’Organisation mondiale de la propriété intellectuelle (OMPI) a publié l’édition 2026 de son Rapport sur la propriété intellectuelle dans le monde, intitulé « La technologie en mouvement », le mardi 17 février à Genève et en ligne. Ce rapport analyse la manière dont les technologies se diffusent à l’échelle mondiale et leurs implications pour le développement économique. Il révèle une accélération spectaculaire de la diffusion mondiale des technologies : alors que les technologies plus anciennes, telles que le télégraphe et l’automobile, ont mis des décennies à se diffuser, les innovations numériques contemporaines, telles que l’IA générative, atteignent les utilisateurs du monde entier en quelques jours grâce à une infrastructure numérique mondiale mature. Les écarts d’adoption entre les économies avancées et les économies en développement se sont réduits pour les technologies récentes, et les différences d’intensité d’utilisation s’amenuisent, en particulier pour les technologies numériques. Toutefois, des disparités importantes subsistent, notamment en Afrique, où les écarts en matière d’infrastructures et d’accès persistent. Le leadership en matière d’innovation reste concentré dans une poignée d’économies, notamment les États-Unis, l’Europe occidentale, le Japon et la Chine. Le succès de la diffusion dépend de quatre facteurs clés : les caractéristiques technologiques, les flux d’information, la capacité d’absorption, ainsi que les politiques publiques et les cadres de propriété intellectuelle. Le rapport souligne que des politiques et des investissements délibérés sont essentiels pour traduire cette diffusion rapide en un développement et une croissance économiques inclusifs.

Sommet sur l’IA à Genève : dix manières dont la Suisse peut contribuer à l’IA et à l’humanité

En 2027, Genève accueillera le prochain sommet mondial sur l’IA, à un moment où les gouvernements, les entreprises et les communautés du monde entier sont profondément engagés dans une transformation axée sur l’IA.

Les hôtes précédents ont apporté leur propre touche distinctive : de l’accent mis par Bletchley Park sur les risques existentiels à l’équilibre entre innovation et sécurité de Séoul, en passant par la perspective économique et sociétale de Paris et l’accent mis par New Delhi sur le développement et l’inclusion. La Suisse a désormais l’opportunité de façonner la prochaine phase de la gouvernance de l’IA et de faire en sorte que le sommet sur l’IA de 2027 soit plus qu’un simple événement.

Nous proposons dix étapes à suivre sur la route menant au sommet sur l’IA de Genève en 2027, soutenues par les recherches, les formations et le suivi des politiques de DiploAI via le Digital Watch Observatory.

Road, Path, City, Field, Outdoors, Nature, Sign, Symbol, Neighborhood, Architecture, Building, Factory

Innovation. L’IA est fondamentalement une question d’innovation, tant technologique que, de plus en plus, sociétale. La prochaine vague d’innovation consistera à activer les connaissances des citoyens et des institutions grâce à l’étiquetage des données, à l’intégration de l’apprentissage par renforcement dans les pratiques pédagogiques et au développement de graphes de connaissances. La Suisse figure depuis longtemps parmi les pays les plus innovants au monde, privilégiant les développements concrets et peu médiatisés qui répondent à des besoins réels et explorent des niches inexploitées.

Gouvernance. Les cadres de gouvernance internationaux existants sont susceptibles d’influencer la politique en matière d’IA à Genève, compte tenu de la concentration dans cette ville d’organisations actives dans les domaines du commerce, de la santé, des télécommunications, du travail et de la sécurité. Le nouveau Groupe scientifique international sur l’IA peut tirer des enseignements de l’expérience du GIEC, basé à Genève, à l’interface entre science et diplomatie. Le modèle suisse d’élaboration des politiques par la base favorise l’inclusion des citoyens dans les débats sur l’IA, tandis que sa tradition réglementaire prudente et fondée sur les lacunes s’aligne sur les appels émergents en faveur d’une gouvernance pragmatique et proportionnée de l’IA.

Subsidiarité. Le principe de subsidiarité, au cœur de l’organisation sociale suisse, stipule que la prise de décision doit se faire au plus près des citoyens et des communautés concernés. Appliquée à l’IA, cette approche permettrait de contrer la concentration du pouvoir entre les mains de quelques grandes plateformes en ancrant le développement de l’IA dans les communautés locales, où les connaissances sont créées à travers les interactions quotidiennes.

EspriTech. L’IA suscite une réflexion renouvelée sur des questions fondamentales liées à l’humanité, au libre arbitre et à l’éthique, amenant les sociétés à revisiter leurs fondements culturels, religieux et philosophiques. S’appuyant sur EspriTech et l’héritage intellectuel des penseurs genevois, ces enseignements peuvent contribuer à affiner le débat sur l’IA et l’humanité.

Confiance. La Suisse, en tant que pays jouissant d’un capital confiance élevé, peut promouvoir une approche « faire confiance mais vérifier » en prévision du sommet de 2027. La confiance peut être rétablie grâce à une discussion réaliste et pleinement informée sur les risques liés à l’IA, qui ont progressivement évolué depuis 2023, passant de l’accent mis sur les risques existentiels (survie de l’humanité) à la primauté actuelle des risques existants (éducation, désinformation, emplois) et aux préoccupations croissantes concernant les risques d’exclusion (monopolisation de l’IA par quelques acteurs).

Apprenticeship. Le modèle d’apprentissage de l’IA, inspiré de la tradition suisse de l’apprentissage par la pratique avec mentorat, s’impose comme un moyen efficace de se former à l’IA. En prévision du Sommet de Genève sur l’impact de l’IA de 2027, il peut renforcer les connaissances et les capacités des diplomates, de la société civile et des communautés locales en matière d’IA.

Humanité. Le Sommet 2027 sur l’IA doit donner un sens concret à l’appel lancé en faveur d’une IA au service des intérêts fondamentaux de l’humanité. La Suisse a toujours mis en pratique son principe d’une société centrée sur l’humain dans les domaines de la politique, de l’éducation, de la protection sociale et de l’économie. Cette expérience séculaire peut contribuer à affiner les liens essentiels entre l’IA et la civilisation humaine, à la fois en partageant certaines leçons apprises et en expérimentant de nouvelles approches et pratiques pour l’ère de l’IA.

Institutions. Les institutions sont des vecteurs importants de la mémoire et des connaissances sociétales. L’IA doit être considérée comme un agent de changement créatif qui peut, entre autres, préserver la mémoire institutionnelle et renforcer la capacité à répondre aux besoins sociétaux.

Multilatéralisme. Le Sommet sur l’IA peut contribuer à clarifier le rôle des organisations internationales à l’ère de l’IA. L’IA peut, par exemple, contribuer à renforcer la légitimité des processus internationaux en garantissant que les contributions aux consultations publiques soient correctement suivies et prises en compte dans les documents politiques.

Souveraineté. Alors que les tensions géopolitiques s’intensifient, les questions relatives à l’IA et à la souveraineté numérique prennent de plus en plus d’importance, soulignant la nécessité d’une action, d’une autodétermination et d’une gestion responsable des connaissances qui sous-tendent l’IA plutôt que l’isolement. À l’approche du sommet de 2027, l’expérience suisse peut contribuer à des discussions plus éclairées sur les dimensions techniques, juridiques et cognitives de la souveraineté en matière d’IA dans un cadre interdépendant.

Veuillez consulter l’intégralité des suggestions sur notre page web dédiée.

Weekly #252 When AI ethics collide with national security: The Anthropic-Pentagon standoff

27 February – 3 March 2026

HIGHLIGHT OF THE WEEK

When AI ethics collide with national security: The Anthropic-Pentagon standoff

In a high-stakes showdown that is redrawing the battle lines between Silicon Valley and the US military, AI firm Anthropic found itself exiled from the Pentagon after refusing to waive its ethical safeguards.

How it started: A contractual dispute. The Pentagon sought assurances that Anthropic’s model, Claude, could be used for ‘all lawful purposes.’ Anthropic pushed back, arguing that such wording did not sufficiently restrict uses the company considers high-risk, particularly mass domestic surveillance and fully autonomous weapons systems. The company requested clearer guardrails.

How it escalated: A supply chain risk designation. Officials signalled that Anthropic risked losing federal contracts and could even face designation as a supply chain risk. The administration ultimately moved ahead with that step, ordering agencies to halt use of Anthropic’s systems and providing a limited six-month wind-down period for existing arrangements.

Supply-chain risk designations are typically for foreign-adversary threats, not for domestic firms negotiating contract terms. Experts and Anthropic argue that such a designation has no clear precedent, and Anthropic plans to challenge it in court.

Enter: The rival company. Into this vacuum stepped OpenAI. Shortly after Anthropic’s blacklisting, OpenAI reached its own arrangement with the Pentagon. The company publicly emphasised that it maintains safety ‘red lines,’ including restrictions related to mass surveillance and the requirement for human oversight in the use of force. CEO Sam Altman indicated that OpenAI shares many of the same ethical concerns Anthropic had raised.

However, Altman ultimately called OpenAI’s contract with the Pentagon rushed, and added on 3 March that the deal would be amended to ensure that the AI system wouldn’t be used for mass domestic surveillance.

The reactions. The episode reverberated throughout the tech world. Major industry groups, including representatives from Amazon, Nvidia, Apple, and others, warned the government against broad use of supply-chain risk designations for US tech companies, fearing chilling effects on innovation and public-private cooperation.

Investors in Anthropic also pushed for de-escalation, worried that the standoff could harm the company’s enterprise business and IPO prospects if key contracts were lost.

However, the day after the Anthropic lost its Pentagon contract, Claude hit number 1 in US app downloads, while US app uninstalls of ChatGPT’s mobile app jumped 295% day-over-day. This suggests that Claude may not lose popularity with users. Additionally, the supply chain risk designation may mean that businesses can work with Anthropic on non-defence-related projects.

Why does it matter? Ultimately, the dispute exposed a deeper structural tension. Advanced AI systems are increasingly central to military planning, logistics, intelligence analysis, and battlefield decision-making. At the same time, leading AI firms have articulated ethical boundaries around surveillance, lethal autonomy, and dual-use risks.

The confrontation between Anthropic and the Pentagon crystallised the question of who determines those boundaries when national security and corporate governance collide.

IN OTHER NEWS LAST WEEK

This week in AI governance

The UN. At the Independent International Scientific Panel on Artificial Intelligence (AI), UN Secretary General Antonio Guterres told experts that they have a huge responsibility to help shape how the technology is used ‘for the benefit of humanity’. As AI development accelerates, the Secretary-General also warned the panel that it is ‘in a race against time.’ Guterres also pointed to earlier work through the UN High-Level Advisory Body on AI, noting that the new scientific panel does not ‘start from zero’. Concluding his remarks, Guterres told the experts: ‘I can think of no more important assignment for our world today.’

UNESCO has signed a cooperation agreement with the National Centre for Artificial Intelligence (CENIA) to promote ethical AI development and strengthen AI education across Chile and Latin America. The partnership will focus on building digital skills, improving AI literacy, and supporting people-centred AI governance through training programmes, educational resources, and multistakeholder collaboration.

The USA. Seven tech giants ( Google, Meta, Microsoft, Oracle, OpenAI, Amazon, and xAI) have signed the White House’s ratepayer protection pledge, encouraging major technology companies to cover the additional electricity costs associated with their AI infrastructure. Participating companies have agreed to finance new power generation resources, upgrade electricity delivery infrastructure and negotiate separate electricity rate structures with utilities and state authorities. The arrangement is designed to ensure that additional energy demand from large data centres does not translate into higher prices for residential consumers.

South Korea and Singapore. South Korea and Singapore have launched a Korea-Singapore AI Alliance, a bilateral initiative to deepen cooperation in AI and related technologies. Announced at the Korea-Singapore AI Connect Summit in Singapore, the alliance aims to create an open innovation ecosystem that connects capital, talent and technology between the two nations, with the goal of enhancing competitiveness in the global AI market and supporting joint development of AI solutions that address regional and global challenges. The partnership includes a pledge to establish a US$300 million global AI investment fund in Singapore by 2030 to support startups and collaborative research.

Child safety online

Yet another country has decided to restrict minors’ access to social media—Indonesia’s Communication and Digital Affairs Minister signed a government regulation that means children under 16 can no longer have accounts on high-risk digital platforms. This will reportedly include YouTube, TikTok, Facebook, Instagram, Threads, X, Bigo Live and Roblox. Implementation of the regulation will begin gradually from 28 March.

The UK is still debating. The country has launched a public consultation examining whether children under 16 should face restrictions or a potential ban on social media use. Young people, parents and educators are being invited to share views before ministers decide on future policy. Potential measures could include setting a minimum age limit for social media, restricting harmful features such as infinite scrolling, and examining protections against children sending or receiving explicit images. The consultation will also explore restrictions on children’s use of AI chatbots and limits on VPN use where it undermines safety protections. The government intends to act swiftly on its findings within months by introducing targeted legal powers that can be enacted rapidly as technology evolves.

Meanwhile, China’s draft measures that classify and regulate online information that may harm minors’ physical and mental health went into effect on 1 March. Covered content includes material inducing unsafe behaviour, extreme emotions, discrimination, unhealthy lifestyles, irrational consumption, celebrity worship, or distorted values such as hedonism and pseudoscience. The rules also restrict the misuse of minors’ images and personal data. Providers of algorithmic recommendation systems and generative AI services are required to strengthen and refine their security governance frameworks and technical safeguards, and are prohibited from promoting or distributing online content that could negatively impact minors’ physical or psychological well-being.

A procedural setback in the European Parliament has cast uncertainty over plans to extend temporary EU rules allowing tech companies to voluntarily scan their services for child sexual abuse material (CSAM). The temporary regime, introduced as a stopgap while the EU negotiates the Child Sexual Abuse Regulation, is due to expire in April 2026. The European Commission has proposed extending it by two years, though some MEPs had pushed to limit the extension to one year. On Monday, Parliament’s civil liberties committee voted on amendments to a draft report on extending the current derogation from EU privacy rules, but ultimately failed to adopt the final text, sending the file to a plenary vote in the week of 9–12 March. Delays now tighten the timeline for reaching an agreement before the deadline.

A few days later, European Commission President Ursula von der Leyen convened the first meeting of the Special Panel on child safety online, announced in her 2025 State of the Union address. The panel will provide expert guidance on protecting and empowering children online and explore potential harmonised age limits for social media access. The inaugural session examined online risks and benefits, tech companies’ responsibilities, addictive features, and digital literacy. The panel will consider the advantages for minors in accessing online spaces at its next meeting. The panel aims to present a report with recommendations to the Commission President by summer 2026.

Why does it matter? These developments are part of a surging global movement to protect children online—one that gained critical momentum with Australia’s push for a social media ban and is now reverberating through policy debates across continents.

Cyber operations intensify in the USA-Israel-Iran conflict

Last weekend, the USA and Israel initiated coordinated military strikes against Iran. According to Chairman of the Joint Chiefs of Staff Gen. Dan Caine, cyber operations conducted by USCYBERCOM have accompanied the strikes. The cyber operations, along with space operations, disrupted Iran’s communications.

Security researchers report that Iranian-aligned actors are preparing potential retaliatory cyberattacks, including DDoS campaigns, ransomware, and destructive malware targeting critical infrastructure and cloud services.

Researchers also observed Iranian-linked attempts to hack IP cameras across the Middle East, potentially to support missile targeting and battle-damage assessment during military operations.

At the same time, Iran is experiencing a near-total internet blackout, possibly a mix of state-imposed restrictions and external cyber disruption.

Why does it matter? The developments highlight how cyber operations are increasingly integrated into interstate conflict. These dynamics are closely linked to ongoing international discussions on responsible state behaviour in cyberspace at the UN, which focus on preventing escalation in cyberspace. The UN’s permanent negotiating forum on cybersecurity, the Global Mechanism on ICT Security, is set to meet for the first time in the last week of March. It remains to be seen how this episode will shape those discussions.

Google cuts Play Store fees after Epic Games settlement

Google is planning major changes to its Play Store policies after settling a long-running legal dispute with Epic Games, the developer behind the popular game Fortnite.

The agreement will reduce the commission Google charges on in-app purchases and introduce new options that make it easier for users to install alternative app stores on Android devices.

Under the new structure, Google will lower its standard commission to 20% on in-app purchases. Developers who choose to use Google’s billing system will pay an additional 5% fee. The company also announced that recurring subscription fees will drop to 10%.

As part of the agreement, Epic Games plans to bring Fortnite back to the Google Play Store globally while continuing to develop its own Epic Games Store for Android.

in processing times, fewer redundancies, and a more transparent user experience. The initiative marks a significant step in Cabo Verde’s digital transformation and modernisation of public administration.

View all updates

LAST WEEK IN GENEVA

GGE on LAWS ninth meeting

The Group of Governmental Experts on Emerging Technologies in the Area of Lethal Autonomous Weapons Systems (GGE on LAWS) met in Geneva this week. Its mandate provides for 10 days of deliberations focused on further consideration and, by consensus, formulation of a set of elements for an instrument—without prejudging its nature—and other possible measures to address emerging technologies in the area of lethal autonomous weapons systems (LAWS). The group’s first meeting of 2026 will wrap up today (6 March). The group will meet once again from 31 August–4 September 2026.

Fifth meeting of the UN CSTD multi-stakeholder working group on data governance at all levels

The UN Commission on Science and Technology for Development (CSTD) held the fifth meeting of its Multi-Stakeholder Working Group on Data Governance at All Levels on 2-3 March 2026 at the Palais des Nations in Geneva, with participation available both in person and online. The two-day meeting focused primarily on developing the report’s structure and content. Discussions included the approval of the meeting agenda, the presentation and review of a preliminary outline of the progress report, and deliberations on synthesis notes prepared by the secretariat and co-facilitators. Participants also considered linkages between different thematic tracks and the outline and annotations of the report.

Security and governance of open source software (OSS): Geneva Dialogue Masterclass #1

The Geneva Dialogue masterclass dedicated to the security and governance of open source software opened the first thematic cycle of 2026 on 4 March 2026. This masterclass demonstrated that open source security requires a comprehensive approach encompassing governance, regulation, collaboration, and geopolitical awareness. The discussion revealed both the fragility and resilience of open source ecosystems, highlighting how sophisticated attacks can exploit vulnerabilities whilst also showing how community collaboration and appropriate regulatory frameworks can strengthen security.

The convergence of AI democratisation, geopolitical tensions, and increasingly sophisticated supply chain attacks creates an environment where traditional security approaches are insufficient. Success will require new forms of collaboration among manufacturers, maintainers, governments, and security researchers, supported by regulatory frameworks that understand and support open-source ecosystems.

LOOKING AHEAD

ICANN85 Community Forum

ICANN85 Community Forum, organised by the Internet Corporation for Assigned Names and Numbers (ICANN), is scheduled to take place from 7 to 12 March 2026 in Mumbai, India. The programme is expected to encompass a range of sessions in which participants engage in policy development activities, advance ongoing technical and operational work, exchange best practices, and discuss matters related to the coordination of the global Domain Name System (DNS) and other unique identifiers essential to internet functionality.

Data Technology Seminar 2026

The Data Technology Seminar 2026, organised by the European Broadcasting Union, will take place from 10 to 12 March in Geneva. The event will bring together media professionals and technology experts to discuss how artificial intelligence and data systems are being developed, governed, and deployed in public service media. Sessions will explore topics such as AI strategy and governance, metadata platforms, hybrid search, audience personalisation, and the use of generative AI in editorial and production workflows.

READING CORNER

BLOG featured image 2026 26 Digital sovereignty stack

Digital sovereignty stack: Infrastructure, services, data, and AI knowledge – Diplo

Definition and analysis of digital sovereignty, consisting of infrastructure, service, data, and knowledge

Social media addiction trial Monthly newsletter March 2026

Digital Watch newsletter – Issue 107 – February 2026

In our February 2026 issue, child safety online takes centre stage as landmark US platform trials begin and countries weigh social media bans for minors. We examine why the “cyberspace is a separate world” myth still shapes AI governance today, plus Anthropic’s Pentagon exile over ethical safeguards. Digital sovereignty trends intensify as Europe pursues strategic autonomy and Washington pushes back. We also preview ten key signposts on the road to the 2027 Geneva AI Summit.

Digital Watch newsletter – Issue 107 – February 2026

February 2026 in retrospect

Child safety online is in the spotlight, as the LA addiction trial and Santa Fe consumer protection trial kick off, marking the first time social media platforms are defending themselves before a jury. Meanwhile, more countries are considering bans on children’s access to social media.

This month’s highlights:

Road to Geneva 2027: Ten key signposts ahead of the next global AI Summit, powered by DiploAI research.

Why cyberspace doesn’t exist: 30 years after the Declaration of the Independence of Cyberspace, we examine how the cyberspace as a separate world myth shapes AI governance today.

Anthropic vs. Pentagon: AI firm Anthropic was barred from Pentagon work after refusing to waive its ethical safeguards—a major Silicon Valley-military showdown.

Tech sovereignty: From European strategic autonomy to US pushback, digital sovereignty continues to dominate 2026.

Snapshot: The developments that made waves in February

Technologies

In the Netherlands, a Dutch appeals court ordered a full investigation into Nexperia and upheld earlier decisions suspending its former CEO, linked to the Chinese parent Wingtech. The court’s ruling strengthens oversight of the company’s governance and operations, reflecting broader European concerns about foreign ownership and control of strategically sensitive semiconductor assets, particularly for the automotive and technology sectors. Nexperia has pledged full cooperation with the probe, while Wingtech criticised the decision as harmful to the global industry.

Taiwan Semiconductor Manufacturing Company (TSMC) announced it will produce advanced 3-nanometre AI chips at its second fabrication plant in Kumamoto, Japan. The expansion underscores surging global demand for AI processors and aligns with Japan’s strategy to strengthen domestic semiconductor capabilities and diversify critical production outside traditional hubs.

The UK and Bulgaria have agreed to deepen cooperation on semiconductors, with the UK’s Science and Technology Network and Department for Business and Trade linking British compound-semiconductor expertise to Bulgaria’s manufacturing base; the partnership’s headline outcome is progress toward a €350 million “Green Silicon Carbide” wafer factory in Bulgaria, alongside new R&D and industry tie-ups including a research memorandum between the Universities of Glasgow and Sofia and an MoU between TechWorks UK and Bulgaria’s BASEL, as both sides frame the push as strengthening European supply-chain resilience and skills.

A US national-security review is holding up licences for Nvidia to ship its H200 AI chips to China, leaving Chinese customers unable to place or confirm orders nearly two months after the White House signalled approval in principle; while the Commerce Department has completed its assessment, inter-agency consultations, including State, Defence and Energy, are still negotiating additional safeguards and potential conditions such as shipment allocation, testing and end-use reporting, delays that are already disrupting demand expectations and production planning across Nvidia’s supply chain and pushing Chinese firms to explore alternative ways to secure AI chips.

Read about this month’s AI governance developments in our dedicated AI newsletter section.

Infrastructure

Engineers are retrieving TAT‑8, the first fibre-optic transatlantic cable, from the Atlantic seabed more than three decades after it revolutionised global communications. Though retired in 2002 after an irreparable fault, the cable has remained submerged until now. The operation clears the seabed for new infrastructure and recovers glass fibre, copper, and steel components for recycling amid global metal shortages.

SpaceX’s Starlink has received regulatory approval to operate in Vietnam, expanding its global footprint and reinforcing its role as an alternative connectivity provider in tightly regulated markets. In parallel, a Russian official acknowledged that Starlink systems had been down for two weeks in parts of Russia.

A coalition of major technology companies announced the creation of the Trusted Tech Alliance (TTA) and introduced five core principles to define what constitutes ‘trusted’ digital infrastructure: transparent corporate governance, secure development and independent assessment, supply chain oversight, ecosystem openness, and adherence to the rule of law and data protection standards. The initiative positions itself as a response to rising geopolitical fragmentation and growing scrutiny over the security of critical digital systems.

Cybersecurity

The Ministry of Public Security of the People’s Republic of China has drafted a law allowing authorities to impose exit bans of up to three years on convicted cybercriminals and those who support or facilitate such activities. It would also bar entry to offenders, extend jurisdiction over Chinese nationals abroad, target foreign entities deemed to harm national interests, and tighten controls on online content deemed false or disruptive. The proposal could affect global businesses, cross-border cooperation, and the international mobility of technology professionals.

The USA and Israel initiated coordinated military strikes against Iran; cyber operations conducted by USCYBERCOM have accompanied the strikes. The cyber operations, along with space operations, disrupted Iran’s communications. Iran’s cyber response has so far been limited to proxy-driven distributed denial-of-service attacks, GPS spoofing near the Strait of Hormuz, and the compromise of IP cameras to support missile operations, alongside a nationwide internet blackout.

The European Commission has launched ProtectEU, a new counterterrorism agenda that sharpens the bloc’s response to evolving threats, especially those amplified by digital tools, by boosting intelligence analysis and Europol support, tightening cooperation with platforms to remove extremist content faster, and stepping up enforcement of the Digital Services Act, while also proposing an the EU Online Crisis Response Framework to coordinate with tech companies during security incidents and expanding measures to protect public spaces, critical infrastructure and disrupt terrorist financing, including via crypto-assets.

Read about this month’s child safety developments in our dedicated child safety newsletter section.

Economic

The European Commission and the COMESA Competition and Consumer Commission are separately investigating concerns that Meta of abusing a dominant position by restricting third-party AI assistants’ access to WhatsApp while privileging its own Meta AI. The European Commission has already formally notified Meta that it has breached EU competition law and is considering interim measures to prevent continued exclusion and protect competitive entry.

France is intensifying its stance against ultra-low-cost online retailers, with Minister Serge Papin declaring 2026 a ‘year of resistance’ to platforms such as Shein. The government argues that global marketplaces benefit from looser regulatory standards than physical French shops. Paris is appealing a court decision that allowed Shein to continue operating despite inappropriate products and is preparing legislation to let authorities suspend online platforms without prior judicial approval, expanding executive powers over the digital economy.

Türkiye’s ruling AK Party has introduced a draft bill to formalise crypto taxation by tying digital-asset rules to the Capital Markets Law and requiring licensed platforms to withhold a 10% tax on crypto gains and income every quarter for individuals and companies, including residents and non-residents. The proposal also adds a 0.03% transaction tax on crypto service providers, obliges investors using unlicensed platforms to declare gains annually, and would let the president adjust the withholding rate from 0% to 20% based on factors such as token type or holding period, with the new taxation regime set to take effect two months after publication if the bill passes.

South Korea’s finance minister Koo Yun-cheol has pledged urgent reforms to how government agencies handle seized and state-held crypto after multiple custody failures, including a case in which Seoul police reportedly lost access to 22 BTC (about $1.4 million) when private keys were not properly retained and a third party was allowed to manage the assets. Prosecutors are also investigating alleged bribery linked to the incident, and the finance ministry says the government holds crypto only through lawful enforcement actions such as seizures in tax and criminal cases.

Russia’s central bank says it is intensifying its crackdown on crypto-enabled pyramid schemes, reporting that two-thirds of such operators now rely on cryptocurrency and that victims’ funds were routed to more than 4,600 fraudster-controlled wallets in 2025. The regulator says it identified 7,087 online scams last year, blocked 21,500 scam-linked webpages and social posts, and is urging Russians to use only licensed investment providers as authorities tighten oversight of online fraud spread via social media, chat apps and phone calls.

Human rights

The EU has dropped plans to revise the GDPR’s definition of ‘personal data’ from the draft GDPR omnibus package after strong pushback from national regulators and civil society, opting to keep the regulation’s current scope intact. Attention now shifts to upcoming European Data Protection Board guidance on pseudonymisation, which is expected to clarify how key safeguards should be applied in practice, signalling a broader preference for regulatory clarity and implementation guidance over reopening foundational privacy concepts in legislation.

Negotiations between Australia and the USA over expanded biometric data sharing have raised alarm among privacy advocates and legal commentators. Reports suggest that discussions could broaden US access to sensitive Australian biometric records — including facial images, fingerprints, and identity data — bypassing traditional case-by-case legal cooperation frameworks.

Italy’s privacy watchdog has ordered Amazon Italia Logistics to stop processing sensitive employee data at its Passo Corese site and to halt the use of data collected via surveillance cameras near restrooms and break areas, after finding the company recorded information such as workers’ health conditions, union/strike activity, and private family details and retained it for up to 10 years, far beyond what authorities say employers may lawfully gather for workplace management.

Nigeria’s Data Protection Commission (NDPC) has launched an inquiry into the Chinese e-commerce giant Temu over suspected violations of Nigeria’s data protection law. Authorities are probing the company’s data-handling practices, specifically, alleged non-transparent data processing, intrusive surveillance mechanisms, cross-border transfers, and possible failure to limit data collection. Temu has pledged cooperation as regulators warn that non-compliance could trigger legal penalties and set precedents for data governance in Africa’s largest digital market.

A draft Law on Cognitive Sovereignty and Protection of Human Attention has been introduced in the Argentine Chamber of Deputies. It suggests establishing a regulatory framework recognising cognitive autonomy as a legally protected good under constitutional and international human rights law, requiring transparency, opt-in personalisation, and non-algorithmic alternatives on mass-reach platforms. It also suggests imposing strict “safe by default” settings for minors, banning behavioural profiling and targeted advertising for users under 13, and mandating impact assessments, registry obligations, audits, and annual transparency reporting.

Sociocultural

UNESCO and Hamad Bin Khalifa University (HBKU) have launched a UNESCO Chair on Digital Technologies and Human Behaviour in Qatar to research how emerging technologies shape daily life, with a focus on digital well-being, ethical design, and healthier online environments; the programme will tackle issues such as internet addiction, cyberbullying, and misinformation, and aims to link research with policy dialogue among governments, international organisations, and academia to promote more responsible technology development.

The UK is introducing legislation requiring tech companies to remove non-consensual intimate images within 48 hours of being reported. Under the updated Crime and Policing Bill, firms that fail to comply risk fines of up to 10% of global revenue or potential service restrictions, with enforcement overseen by Ofcom.

The US Department of State is reportedly preparing to launch ‘freedom.gov,’ an online portal designed to help users worldwide, including in Europe and elsewhere, circumvent local content restrictions and access blocked material, including content their governments classify as hate speech or terrorist propaganda.

The EU is investigating whether Shein’s design elements, such as gamified engagement and opaque recommendation algorithms, undermine consumer safety and transparency obligations under the DSA. Investigators will examine whether Shein has failed to prevent the sale of illegal products — including items that may constitute child sexual abuse material. Shein’s risk-mitigation systems, product removal processes, and compliance with requirements to offer non-profiling recommendation options will be evaluated, with potential fines of up to 6 % of global turnover for confirmed breaches.

The European Commission has preliminarily concluded that TikTok’s design violates the bloc’s Digital Services Act (DSA) due to features that the Commission considers addictive, such as infinite scroll, autoplay, push notifications, and its highly personalised recommender system. According to the Commission, existing safeguards on TikTok—such as screen-time management and parental control tools—do not appear sufficient to mitigate the risks associated with these design choices. At this stage, the Commission indicates that TikTok would need to modify the core design of its service. Possible measures include phasing out or limiting infinite scroll, introducing more effective screen-time breaks, including at night, and adjusting its recommender system to reduce addictive effects.

Development

Malaysia has imposed an immediate, total ban on all e-waste imports, reclassifying electronic waste under an ‘absolute prohibition’ in its import rules, after a widening corruption investigation into oversight of the sector. Authorities say the move is meant to stop foreign dumping and protect public health and national security, while warning that enforcement will be tightened to prevent smuggling. The probe has reportedly led to the detention of senior environmental officials and asset freezes.

Gabon has imposed an indefinite suspension of social media platforms, citing the spread of false information, cyberbullying and the unauthorised disclosure of personal data. Gabon’s media regulator, the High Authority for Communication (HAC), stated that existing moderation measures were not working and that the shutdown was necessary to stop violations of Gabon’s 2016 Communications Code.

The European Commission has proposed opening negotiations to bring Albania, Bosnia and Herzegovina, Kosovo, Montenegro, North Macedonia, and Serbia into the EU’s ‘Roam Like at Home’ regime. If implemented, citizens and businesses would be able to make calls, send texts, and use mobile data across borders at domestic rates, both when visiting the EU and when EU citizens travel in the region. The Commission has adopted proposals for negotiating mandates and is now seeking approval from the European Council to begin formal talks.

The government of Cabo Verde has launched Gov.CV, a unified digital portal designed to centralise public services and streamline interactions between the state, citizens, and businesses. By consolidating services, the government expects reductions in processing times, fewer redundancies, and a more transparent user experience.

AI governance in February

Global governance

The UN. The General Assembly approved the creation of a historic global scientific advisory body on AI, the Independent International Scientific Panel on Artificial Intelligence (AI), tasked with providing independent, evidence‑based assessments of AI technologies, risks, opportunities, and impacts.

The first of its kind, the panel’s main task is to ‘issuing evidence-based scientific assessments synthesising and analysing existing research related to the opportunities, risks and impacts of AI’, in the form of one annual ‘policy-relevant but non-prescriptive summary report’ to be presented to the Global Dialogue on AI Governance. The Panel will also ‘provide updates on its work up to twice a year to hear views through an interactive dialogue of the plenary of the General Assembly with the Co-Chairs of the Panel’. At the Panel’s inaugural meeting, Guterres told experts that they have a huge responsibility to help shape how the technology is used ‘for the benefit of humanity’.

India AI Impact Summit 2026. In a first for the Global South, India hosted the world’s biggest AI summit at Bharat Mandapam.

The New Delhi Frontier AI Impact Commitments were unveiled at the summit’s opening, centring on two core priorities. The first, advancing understanding of real-world AI usage, seeks to generate anonymised and aggregated data to inform policymaking on AI’s impact on jobs, skills and productivity. The aim is to support evidence-based regulation and economic planning as AI adoption accelerates.

The second, strengthening multilingual and contextual evaluations, centres on improving AI performance across underrepresented languages and cultural contexts. Participating organisations will collaborate with governments and local ecosystems to develop datasets, benchmarks and evaluation expertise, with a particular emphasis on the Global South.

At the heart of the summit was Prime Minister Narendra Modi’s unveiling of the ‘MANAV Vision’, a human-centred approach to AI governance. Framed as a series of principles aimed at placing people at the centre of AI development and deployment, MANAV stands for:

Moral and ethical systems — ensuring AI is guided by ethical norms
Accountable governance — transparent rules and oversight mechanisms
National sovereignty — rights over data and digital assets
Accessible and inclusive AI — avoiding monopolies and broadening participation
Valid and legitimate systems — lawful and verifiable technologies.

Modi described this framework as essential to preventing future disparities in AI’s impact and ensuring technology serves humanity’s welfare. He also emphasised that AI should be a medium for inclusion and empowerment, particularly for the Global South, rather than a tool that concentrates power among a few actors.

The 2027 edition of the summit will be hosted by Switzerland. Read more about the road to the 2027 summit in our dedicated newsletter text.

Investments and national plans

The USA. Seven tech giants ( Google, Meta, Microsoft, Oracle, OpenAI, Amazon, and xAI) have signed the White House’s ratepayer protection pledge, encouraging major technology companies to cover the additional electricity costs associated with their AI infrastructure and, in some cases, to invest in dedicated energy generation rather than relying solely on the public grid.

Germany. Germany has unveiled plans for a ‘Sovereign AI Factory’, a government‑backed initiative to develop sovereign AI models and infrastructure tailored to local language, cultural context and industrial needs. The project will support domestic innovation by providing compute resources, datasets and certification frameworks that conform to European safety and privacy standards, with the aim of reducing reliance on non‑EU AI providers. Berlin says the factory will also serve as a collaborative platform for research institutions and industry to co‑design secure, interoperable AI systems for public and private sectors.

Pakistan. Pakistan’s government has pledged major investment in AI by 2030, rolling out a comprehensive national strategy to accelerate digital transformation across the economy. The plan focuses on building AI capacity in key sectors — including agriculture, healthcare and education — through funding for research hubs, public‑private partnerships and targeted upskilling programmes. Officials say the investment is intended to attract foreign direct investment, boost exports and position Pakistan as a regional tech player, while also addressing ethical and governance frameworks to guide responsible AI deployment.

Slovenia. Slovenia has set out an ambitious national AI vision, outlining strategic priorities such as human‑centric AI, robust ethical frameworks, and investment in research and talent. The roadmap emphasises collaboration with European partners and adherence to international standards, positioning Slovenia as a proactive voice in shaping AI governance dialogues.

Partnerships

South Korea and Singapore. South Korea and Singapore have launched a Korea-Singapore AI Alliance, a bilateral initiative to deepen cooperation in AI and related technologies. Announced at the Korea-Singapore AI Connect Summit in Singapore, the alliance aims to create an open innovation ecosystem that connects capital, talent and technology between the two nations, with the goal of enhancing competitiveness in the global AI market and supporting joint development of AI solutions that address regional and global challenges. The partnership includes a pledge to establish a US$300 million global AI investment fund in Singapore by 2030 to support startups and collaborative research.

Content governance

China. A court in eastern China has set an early legal precedent by limiting developer liability for AI hallucinations, ruling that developers are not automatically responsible unless users can prove fault and demonstrable harm. Judges characterised AI services as service providers, requiring claimants to show both provider fault and actual injury from erroneous outputs, a framework intended to balance innovation incentives with user protection.

DPAs. Data protection authorities from 61 jurisdictions and the European Data Protection Supervisor (EDPS) issued a joint statement warning about AI tools that generate realistic images of identifiable individuals without consent. They raised concerns about privacy, dignity, and child safety, noting that such technologies—often embedded in social media—enable non-consensual intimate imagery and other harmful content. Authorities stressed that AI systems must comply with data protection laws and that certain uses may constitute criminal offences. Organisations were urged to implement safeguards, ensure transparency, enable swift content removal, and engage proactively with regulators to protect fundamental rights.

India. India has begun enforcing a three-hour removal rule for AI-generated deepfake content, requiring platforms and intermediaries to take down specified material within 180 minutes of notification or face regulatory sanctions. The accelerated timeframe is designed to blunt the rapid spread of deceptive, synthetic media amid heightened concerns about misinformation and social disruption.

Global coalition on child safety. A broad coalition of child rights advocates, digital safety organisations and policymakers has called on governments to ban ‘nudification’ AI tools, urging criminalisation of software that converts clothed images into sexually explicit versions without consent. The group argues that existing content moderation approaches are insufficient to protect minors and stresses that pre-emptive legal prohibitions are needed to prevent widespread exploitation.

UNICEF. The UN Children’s Fund (UNICEF) has called on governments to criminalise the creation, possession and distribution of AI-generated child sexual abuse content, warning of a sharp rise in sexually explicit deepfakes involving children and urging stronger safety-by-design practices and robust content moderation. A study cited by the agency found that at least 1.2 million children in 11 countries reported their images being manipulated into explicit AI deepfakes, with ‘nudification’ tools that strip or alter clothing posing heightened risks. UNICEF stressed that sexualised deepfakes of minors should be treated as child sexual abuse material under the law and urged digital platforms to prevent circulation rather than merely remove content after the fact.

Spain. In Spain, Prime Minister Pedro Sánchez has ordered prosecutors to investigate X, Meta, and TikTok over the alleged circulation of AI-generated child sexual abuse material (CSAM). The probe follows reports that platform systems may have enabled the creation and spread of sexually explicit deepfake imagery involving minors. Spanish authorities are examining whether companies failed to prevent the distribution of such content and whether AI tools embedded in or linked to the platforms contributed to the harm.

The UK. Britain is partnering with Microsoft, academics, and tech experts to develop a deepfake detection system to combat harmful AI-generated content. The government’s framework will standardise how detection tools are evaluated against real-world threats such as impersonation and sexual exploitation, building on recent legislation criminalising the creation of non-consensual intimate synthetic imagery. Officials cited a dramatic increase in deepfakes shared online in recent years as motivation for the initiative.

Grok/X. The cybercrime unit of the Paris prosecutor has raided the French office of X as part of this expanded investigation. Musk and former CEO Linda Yaccarino have been summoned for voluntary interviews. X denied any wrongdoing and called the raid an ‘abusive act of law enforcement theatre’ while Musk described it as a ‘political attack.’

The UK Information Commissioner’s Office (ICO) opened a formal investigation into X and xAI over whether Grok’s processing of personal data complies with UK data protection law, namely core data protection principles—lawfulness, fairness, and transparency—and whether its design and deployment included sufficient built-in protections to stop the misuse of personal data for creating harmful or manipulated images.

Ireland’s Data Protection Commission (DPC) has initiated a large-scale GDPR investigation into X’s AI chatbot Grok, after reports that its generative AI capabilities have been used to produce harmful, non-consensual and sexualised content involving personal data. This probe, triggered by widespread controversy over Grok’s image outputs, unfolds alongside evidence that the chatbot has been gaining market share in the USA as global regulators scrutinise its compliance with fundamental data protection standards.

Brazil’s National Data Protection Agency and National Consumer Rights Bureau have ordered X to stop serving explicit image generation via its Grok AI, citing risks of harmful outputs reaching minors and contravention of local digital safety norms. The directive demands immediate technical measures to block certain prompts and outputs as part of ongoing scrutiny of platform content moderation practices.

Meanwhile, Indonesia has restored access to Grok after banning it in January, having received guarantees from X that stronger safeguards will be introduced to prevent further misuse of the AI tool.

Chile. Chile has introduced Latam-GPT to strengthen Latin America’s presence in global AI. The project, developed by the National Centre for Artificial Intelligence with support across South America, aims to correct long-standing biases by training systems on the region’s own data instead of material drawn mainly from the USA or Europe. President Gabriel Boric said the model will help maintain cultural identity and allow the region to take a more active role in technological development. Latam-GPT is not designed as a conversational tool but rather as a vast dataset that serves as the foundation for future applications. More than eight terabytes of information have been collected, mainly in Spanish and Portuguese, with plans to add indigenous languages as the project expands.

Safety and security

International experts. The second International AI Safety Report 2026 has been published. The report synthesises evidence on AI capabilities — such as improved reasoning and task performance — alongside emerging risks like deepfakes, cyber misuse and emotional reliance on AI companions, while noting uneven reliability and ongoing challenges in managing risks. It aims to equip policymakers with a science-based foundation for regulatory and governance decisions without prescribing specific policies.

The UN. AI governance was a key focus at the recent UN Special Dialogue entitled ‘From Principles to Practice: Special Dialogue on Artificial Intelligence and Preventing and Countering Violent Extremism’. Diplomats and experts discussed how AI is reshaping global stability, conflict dynamics and international law. Participants highlighted risks from autonomous systems and misinformation campaigns and stressed the need for multilateral cooperation and shared norms to mitigate emerging threats.

The EU. The European Commission has confirmed it will again delay publishing guidance on high-risk AI systems under the EU AI Act. The guidelines were due by 2 February 2026, but will now follow a revised timeline. The delay marks the second missed deadline and adds to broader implementation setbacks surrounding the EU AI Act.

Intellectual property rights

The UK. The UK Supreme Court has ruled that AI-assisted inventions can qualify for patents when the human contributor’s inventive role is identifiable and substantial, a decision legal experts say will boost innovation by clarifying intellectual property protections in hybrid human-AI development. The judgement aims to incentivise investment in AI research while maintaining established patentability standards.

Future of work

South Korea. South Korea has launched a labour‑government body to address the pressures of AI automation on the workforce, creating a cross‑sector council tasked with forecasting trends in job displacement and recommending policy responses. The initiative brings together labour unions, industry leaders and government ministries to coordinate reskilling and upskilling programmes, strengthen social safety nets, and explore income support models for workers affected by automation.

Bellwether trials begin as more countries eye child social media bans

The trials begin

The LA addiction trial. A landmark trial opened in Los Angeles, USA, in February 2026 against Meta and YouTube, centring on claims that their platforms are deliberately designed to be addictive and have harmed young users’ mental health.

The plaintiff, Kaley, 20 years old in 2026, alleges that Instagram and YouTube caused her anxiety, body dysmorphia, and suicidal thoughts. Her lawyers likened features like infinite scroll, autoplay, likes, and beauty filters to a ‘digital casino’ for children, citing internal documents showing the platforms targeted young users and even used YouTube as a ‘digital babysitter.’

Kaley also initially sued Snap Inc. and TikTok, but these platforms reached confidential settlements before trial, leaving Meta and YouTube as the remaining defendants to face a jury.

Meta and YouTube’s defence argued that social media was not responsible for Kaley’s struggles, citing her difficult family background, therapists’ records, and the availability of safety tools.

YouTube highlighted that Kaley’s average daily usage has been 29 minutes since 2020 and compared the platform to other entertainment services, emphasising that she is not addicted.

Meta CEO Mark Zuckerberg took the stand and insisted that Instagram prohibits users under 13 and that enforcing age limits is challenging, as many minors lie about their birth dates. He highlighted ongoing efforts to reduce screen time and improve safety features. Still, internal documents presented in court suggested that early teen engagement had been a strategic priority. The case is being closely watched as a potential blueprint for platform accountability regarding addictive features.

The plaintiff’s psychiatrist, Virginia Burke, also took the stand. Burke testified that plaintiff Kaley’s social media use contributed to her mental health issues, citing online bullying. However, Burke noted that Kaley also enjoyed creating and sharing video art, though she was frustrated when others claimed credit for it. Burke stated that social media addiction is not yet a widely recognised diagnosis in psychiatry and is absent from the latest Diagnostic and Statistical Manual, the key text for US mental health professionals.

Kaley herself testified that from a young age, she spent nearly all her time on platforms such as YouTube and Instagram, describing an inability to limit her use, even when she experienced bullying. She said she secretly retrieved her phone at night after her mother confiscated it, became distressed when denied access to social media apps, withdrew from family interactions, and believed her health, sleep, grades, and overall well-being would have been better without social media.

The trial is expected to last until the end of March 2026.

The Santa Fe consumer protection trial. Another trial opened in Santa Fe, New Mexico, USA, after more than two years of pre-trial wrangling. The lawsuit, filed in 2023 by New Mexico Attorney General Raúl Torrez, has reached the court and a jury.

The lawsuit accuses Meta of violating the state’s consumer protection laws by misrepresenting how safe its platforms are for minors while building features and algorithms that, in prosecutors’ view, entice prolonged use and expose children to significant risks. Those risks include addiction-like engagement, exposure to harmful sexual content, unwanted private communications with adults, sleep disruption from compulsive use, and environments where predators can operate with relative ease.

On 3 March, state attorneys introduced recorded depositions of Meta’s CEO Mark Zuckerberg and Head of Instagram Adam Mosseri, seeking to show that the company was aware of metrics and research indicating serious child safety problems but did not sufficiently act or warn users and families. In those depositions, prosecutors pressed Meta’s executives on issues like safety priorities versus corporate profits, the scope of their platforms’ reach among teens, and specific product choices — from recommendation systems to cosmetic filters — that might affect teen well-being.

Meta’s attorneys told the jury during opening statements that the company has implemented numerous safety tools and content-moderation systems. Meta maintains it has not engaged in deception and that risk disclosures and safety measures have been ongoing.

The Oakland school districts trial. Another bellwether trial is expected to start in Oakland, California, in June, the first to represent school districts that have sued social media platforms over harms to children.

The heart of the matter. Approximately 1,600 plaintiffs — including individuals, more than 350 families, 250 school districts, and state attorneys general — have filed claims against Meta Platforms, Snap Inc., TikTok, and Google. Because of the volume, the lawsuits are coordinated in a Judicial Council Coordination Proceeding (JCCP). From this coordinated group, 22 bellwether trials have been selected, with three — in Los Angeles, Santa Fe, and Oakland — scheduled to proceed first.

A bellwether trial is a test case chosen from a large group of similar lawsuits to be tried first, to gauge how juries respond to the evidence and legal arguments. Its outcome does not decide the other cases, but it signals how future trials or settlement negotiations may unfold.

This is why the Los Angeles and Santa Fe trials have drawn much attention—it is expected that their outcome will influence future platform design practices.

The bans club grows

The momentum to ban children from accessing social media continues, as 10 more nations weigh legislative measures and enforcement tools.

Portugal’s parliament has approved a law restricting social media access for minors under 16, requiring express and verified parental consent for accessing platforms like Instagram, TikTok, and Facebook. Access will be controlled through the Digital Mobile Key, Portugal’s national digital ID system, ensuring effective age verification and platform compliance. The law strengthens protections amid growing concerns over social media’s impact on young people’s mental health, and detailed implementation and enforcement rules are now set for parliamentary committee review.

In Spain, Prime Minister Pedro Sánchez’s government has proposed legislation that would ban social media access for users under 16, framing the measure as a necessary child-protection tool against addiction, exploitation, and harmful content. Under the draft plan, platforms must deploy mandatory age-verification systems designed as enforceable barriers rather than symbolic safeguards—signalling a shift toward stronger regulatory enforcement rather than voluntary compliance by tech companies. Proposals also include legal accountability for technology executives over unlawful or hateful material that remains online.

Türkiye’s ruling AK Party has proposed a bill banning social media access for children under 15, requiring platforms to implement age verification. Platforms will also have to create a separate child-specific version of the platform for older minors aged 15-18. Officials cited the Turkish Penal Code, which limits criminal liability for children under 15, as justification for drawing the line at age 15.

Poland’s ruling coalition is currently drafting a law that would ban social media use for children under 15. Lawmakers aim to finalise the law by late February 2026 and potentially implement it by Christmas 2027. Poland aims to update its digital ID app, mObywatel, to enable users to verify their age.

Slovenia is preparing draft legislation to ban minors under 15 from accessing social media, a move initiated by the Education Ministry.

Greece is reportedly close to announcing a ban on social media use for children under 15. The Ministry of Digital Governance intends to rely on the Kids Wallet application, introduced last year, as a mechanism for enforcing the measure instead of developing a new control framework.

In Austria, the government is actively debating a prohibition on social media use for children under 14. State Secretary for Digital Affairs Alexander Pröll confirmed the policy is under discussion with the aim of bringing it into force by the start of the school year in September 2026.

Germany is weighing limits on children’s access to social media, as the ruling party urges the federal government to introduce a legal minimum age of 14. Chancellor Friedrich Merz has signalled support for the proposal, saying he has considerable sympathy for the idea.

The UK has launched a public consultation examining whether children under 16 should face restrictions or a potential ban on social media use. Young people, parents and educators are being invited to share views before ministers decide on future policy. Potential measures could include setting a minimum age limit for social media, restricting harmful features such as infinite scrolling, and examining protections against children sending or receiving explicit images. The consultation will also explore restrictions on children’s use of AI chatbots and limits on VPN use where it undermines safety protections. The government intends to act swiftly on its findings within months by introducing targeted legal powers that can be enacted rapidly as technology evolves.

The EU as a whole is revisiting the idea of an EU-wide social media age restriction. The issue was raised in the European Commission’s new action plan against cyberbullying, published on Tuesday, 10 February. The plan confirms that a panel of child protection experts will advise the Commission by the summer on possible EU-wide age restrictions for social media use. The panel will assess options for a coordinated European approach, including potential legislation and awareness-raising measures for parents.

The document notes that diverging national rules could lead to uneven protection for children across the bloc. A harmonised EU framework, the Commission argues, would help ensure consistent safeguards and reduce fragmentation in how platforms apply age restrictions.

These individual national efforts unfold against a backdrop of increasing international regulatory coordination. On 3 February 2026, the European Commission convened with Australia’s eSafety Commissioner and the UK’s Ofcom to share insights on age assurance measures—technical and policy approaches for verifying users’ ages and enforcing age‑appropriate restrictions online. The meeting followed a joint communication signed at the end of 2025, where the three regulators pledged ongoing collaboration to strengthen online safety for children, including exploring effective age‑assurance technologies, enforcement strategies, and the role of data and independent research in regulatory action.

The big picture. The membership of the ban club has reached double digits. We’ll continue following the developments.

Zooming out. These initiatives across multiple nations confirm that Australia’s social media ban was not an isolated policy experiment, but rather the beginning of a global bandwagon effect. This momentum is particularly striking given that Australia’s own ban is not yet widely deemed a success—its effectiveness and broader impacts are still being studied and debated.

The developments come just as Australia’s eSafety report notes that tech giants—including Apple, Google, Meta, Microsoft, Discord, Snap, Skype and WhatsApp—have made only limited progress in combating online child sexual exploitation and abuse (CSEA) despite being legally required to report measures under Australia’s Online Safety Act.

From bans to compliance

Beyond outright bans, a second regulatory front is taking shape: tightening the legal and technical conditions in which platforms process minors’ data and host content that may be harmful to children.

Meanwhile, enforcement action in the UK underscores the financial and reputational risks of non-compliance. The UK privacy watchdog fined Reddit £20 million for unlawfully processing children’s personal data and failing to protect under-13 users. The regulator found Reddit lacked ‘robust age assurance mechanisms’ and relied on easily bypassed self-declaration, meaning it had no lawful basis to handle children’s data and exposed them to potentially harmful content. Reddit also did not complete a required data protection impact assessment before 2025. The fine is the largest issued by the ICO over children’s privacy. Reddit plans to appeal.

The data protection authority of Türkiye has opened a new review into how major social media platforms manage children’s personal data. The Personal Data Protection Authority is reviewing how children’s personal data is processed on TikTok, Instagram, Facebook, YouTube, X and Discord and what safeguards are in place. Separately, the ruling Justice and Development Party (AKP) is expected to introduce a family package that would require identity verification for every account through phone numbers or the e-Devlet system. Children under 15 would not be allowed to create profiles, and further limits could apply to users under 18.

In Brazil, a new legislative proposal—Bill No. 730/2026—has been introduced to the Chamber of Deputies. The proposal mandates age-verification mechanisms aligned with Brazil’s data protection law (Law No. 13.709/2018), prioritising data minimisation, pseudonymisation, and limited retention. It also bans direct or indirect monetisation involving children under 14 and requires prior judicial authorisation for remunerated artistic work by adolescents aged 14–18, subject to safeguards. If adopted, the bill would formalise baseline compliance requirements for platforms operating in the Brazilian market.

China’s draft measures that classify and regulate online information that may harm minors’ physical and mental health went into effect on 1 March. Covered content includes material inducing unsafe behaviour, extreme emotions, discrimination, unhealthy lifestyles, irrational consumption, celebrity worship, or distorted values such as hedonism and pseudoscience. The rules also restrict the misuse of minors’ images and personal data. Providers of algorithmic recommendation systems and generative AI services are required to strengthen and refine their security governance frameworks and technical safeguards, and are prohibited from promoting or distributing online content that could negatively impact minors’ physical or psychological well-being.

Why cyberspace doesn’t exist

Thirty years ago, on 8 February 1996, two developments kicked off a powerful narrative about the internet: That it occupied a realm apart from ordinary law and politics. These are the Declaration of the Independence of Cyberspace and the US Communications Decency Act (CDA).

Declaration of the Independence of Cyberspace. In Davos, John Perry Barlow’s Declaration of the Independence of Cyberspace asserted that the ‘Governments of the Industrial World’ have ‘no sovereignty’ in cyberspace.

This vision spawned a generation of thought arguing that the internet meant the ‘end of geography.’ Thousands of articles, books, theses and speeches have been delivered arguing that we need new governance for the ‘new brave world’ of the digital.

This intellectual and policy house of cards was built on the assumption that there is cyberspace beyond physical space. It was (and is) a wrong assumption. There is no cyberspace. Every email, every post, every AI query is ultimately a physical event: pulses of electrons carrying bits and bytes through cables under the ocean, Wi-Fi, data servers, and internet infrastructure.

The CDA and its Section 230. On the same day as Barlow’s declaration, President Clinton signed into law the US Communications Decency Act (CDA), which had been adopted by the US Congress. Buried within it was Section 230, which granted internet platforms an unprecedented immunity: they could not be treated as publishers or speakers of the content they hosted.

For the first time in history, commercial entities were granted a broad shield from liability for the very business from which they profited. It was a departure from the long tradition of legal liability, for example, of a newspaper for the text it publishes or of broadcasters for their transmissions.

This provision was justified as a way to protect a nascent industry from crippling litigation. At the time, internet companies were small and experimental. The immunity enabled rapid growth and innovation.

Over time, however, those start-ups became some of the most valuable corporations in history, with global reach and market capitalisations of trillions of dollars. The legal framework, however, largely remained intact, even as internet companies developed sophisticated algorithms that curate, amplify, and monetise user content at scale. This divergence created a central tension in contemporary law and economics: immensely powerful intermediaries operating with limited accountability for systemic effects.

The convergence of the two. The conceptual separation of ‘cyberspace’ made this arrangement easier to defend. If the internet were a new world, exceptional rules seemed justified.

But critics quickly challenged that reasoning. US judge Frank H. Easterbrook argued that we do not need internet law, as we did not need the ‘law of the horse’ when horses were introduced as the dominant mode of transportation. The internet should be regulated by applying existing legal principles. Law regulates relationships among people and institutions, regardless of the technologies they use. The medium may change; the underlying principles endure.

Experience has largely vindicated that view. Digital technologies have not dissolved geography; they have intensified it. States assert jurisdiction over data flows, content moderation, taxation, competition, and security. High-precision geolocation, data localisation requirements, and national regulatory regimes demonstrate that the internet operates squarely within territorial boundaries.

However, CDA remains in force, extending into the age of AI. Companies developing large language models and other AI systems often rely on intermediary protections and analogous doctrines to limit liability. As a result, AI tools can be deployed globally with comparatively limited ex ante oversight. Yet their outputs can shape public discourse, influence elections, affect mental health, and generate economic disruption.

The central question is not whether innovation should be constrained, but whether it should be aligned with established principles of responsibility. Technologies do not exist outside society; they are embedded within it. If an entity designs, deploys, and profits from a system, it should bear responsibility for its foreseeable impacts. The age of legal exceptionalism should end.

Governments continue the push for digital sovereignty

Last month saw further developments pointing to digital sovereignty as the prevailing trend, carrying over from December 2025 into January and February 2026.

The European Commission has begun testing the open-source Matrix protocol as a possible alternative to proprietary messaging platforms for internal communication. Matrix’s federated architecture allows communications to be hosted on European infrastructure and governed under EU rules, aligning with broader efforts to build sovereign digital public services and reduce reliance on external platforms.

The Commission also unveiled EURO‑3C, a new initiative worth €75 million under the Horizon Europe programme to build Europe’s first large‑scale federated telco‑edge‑cloud infrastructure. By federating existing national infrastructures across borders, EURO‑3C aims to reduce dependence on non‑EU hyperscalers and fortify the EU’s role in cloud, edge computing and AI infrastructure.

In France, the government has taken a hard line on control of satellite infrastructure, another cornerstone of digital sovereignty. Paris blocked the sale of ground-station assets owned by Eutelsat to an external investor, arguing that such infrastructure underpins both civilian and military space communications and must remain under domestic authority. French officials described these facilities as critical to strategic autonomy, in part because Eutelsat represents one of Europe’s few genuine competitors to US-led satellite constellations such as Starlink.

In Russia, the telecommunications regulator Roskomnadzor has tightened restrictions on Telegram, slowing delivery of media and limiting certain features to pressure users toward domestic alternatives. Roskomnadzor stated that Telegram is not taking meaningful measures to combat fraud, is failing to protect users’ personal data, and is violating Russian laws. Telegram’s founder has condemned the measures as authoritarian, warning they may interfere with essential communication services.

This crackdown has escalated with the full blocking of Meta’s WhatsApp, which 100 million Russians use. Authorities justified the ban by pointing to WhatsApp’s refusal to meet Russian legal requirements. Users are being encouraged to adopt government-supported platforms that critics say enable state surveillance, raising concerns about privacy and access to independent communication channels. Meta called the ban harmful to both safety and privacy.

Despite these moves, Russia is pausing aggressive action against Google, citing the country’s dependence on Android devices and warning that a sudden ban could disrupt millions of users. Officials indicated that any transition to domestic alternatives will be gradual, reflecting a cautious approach to reducing reliance on foreign tech.

Meanwhile, in the Netherlands, digital sovereignty has moved to the forefront of parliamentary debate. Lawmakers have renewed calls to shift public and private-sector data away from US-based cloud services, citing risks under US legislation such as the Cloud Act. Concerns have intensified following the proposed acquisition of Solvinity, which hosts parts of the Dutch DigiD digital identity system, by a US firm. MPs emphasised the need for stronger safeguards, the promotion of European or Dutch cloud alternatives, and the updating of procurement rules to protect sensitive data.

As European policymakers weigh strategic autonomy and regulatory control, Washington is simultaneously stepping up efforts to counter what it views as potentially disruptive measures on a global level.

An internal US State Department cable seen by Reuters directs US diplomats to actively oppose foreign data sovereignty and data localisation laws and to promote a more assertive US international data policy. The cable, signed on 18 February, argues that such regulations could disrupt global data flows, raise costs, create unnecessarily burdensome compliance requirements, and hamper cloud and AI services. The directive also encourages advocacy for the Global Cross‑Border Privacy Rules Forum (CBPR) as an alternative mechanism supporting data flow with privacy protections.

Zooming out. This move underscores rising tensions with Europe’s regulatory push around privacy and digital sovereignty and reflects a move toward defending US tech interests abroad.

The big picture. The common thread is clear: Digital sovereignty is now a key consideration for governments worldwide. The approaches may differ, but the goal remains largely the same – to ensure that a nation’s digital future is shaped by its own priorities and rules. But true independence is hampered by deeply embedded global supply chains, prohibitive costs of building parallel systems, and the risk of stifling innovation through isolation. While the strategic push for sovereignty is clear, untangling from interdependent tech ecosystems will require years of investment, migration, and adaptation. The current initiatives mark the beginning of a protracted and challenging transition.

When AI ethics collide with national security: The Anthropic-Pentagon standoff

In a high-stakes showdown that is redrawing the battle lines between Silicon Valley and the US military, AI firm Anthropic found itself exiled from the Pentagon after refusing to waive its ethical safeguards.

How it started: A contractual dispute. The Pentagon sought assurances that Anthropic’s model, Claude, could be used for ‘all lawful purposes.’ Anthropic pushed back, arguing that such wording did not sufficiently restrict uses the company considers high-risk, particularly mass domestic surveillance and fully autonomous weapons systems. The company requested clearer guardrails.

How it escalated: A supply chain risk designation. Officials signalled that Anthropic risked losing federal contracts and could even face designation as a supply chain risk. The administration ultimately moved ahead with that step, ordering agencies to halt use of Anthropic’s systems and providing a limited six-month wind-down period for existing arrangements.

Supply-chain risk designations are typically for foreign-adversary threats, not for domestic firms negotiating contract terms. Experts and Anthropic argue that such a designation has no clear precedent and is likely to be challenged in court as legally unsound.

Enter: The rival company. Into this vacuum stepped OpenAI. Shortly after Anthropic’s blacklisting, OpenAI reached its own arrangement with the Pentagon. The company publicly emphasised that it maintains safety ‘red lines,’ including restrictions related to mass surveillance and the requirement for human oversight in the use of force. CEO Sam Altman indicated that OpenAI shares many of the same ethical concerns Anthropic had raised. The Pentagon accepted OpenAI’s framework, raising questions about why similar safeguards proved untenable in Anthropic’s case.

The reactions. The episode reverberated throughout the tech world. Major industry groups, including representatives from Amazon, Nvidia, Apple, and others, warned the government against broad use of supply-chain risk designations for US tech companies, fearing chilling effects on innovation and public-private cooperation.

Investors in Anthropic also pushed for de-escalation, worried that the standoff could harm the company’s enterprise business and IPO prospects if key contracts were lost.

However, the day after the Anthropic lost its Pentagon contract, Claude hit number 1 in US app downloads, while US app uninstalls of ChatGPT’s mobile app jumped 295% day-over-day. This suggests that Claude may not lose popularity with users.

Why does it matter? Ultimately, the dispute exposed a deeper structural tension. Advanced AI systems are increasingly central to military planning, logistics, intelligence analysis, and battlefield decision-making. At the same time, leading AI firms have articulated ethical boundaries around surveillance, lethal autonomy, and dual-use risks.

The confrontation between Anthropic and the Pentagon crystallised the question of who determines those boundaries when national security and corporate governance collide.

Yet as Anthropic and the Pentagon reportedly return to the negotiating table, it underscores that the issue is far from resolved.

Last month in Geneva

Decoding the UN CSTD Working Group on Data Governance | Part 3

On 10 February, Diplo, the Open Knowledge Foundation, and the Geneva Internet Platform coorganised an online event, ‘Decoding the UN CSTD Working Group on Data Governance | Part 3’, which reviewed progress and prospects of the UN Multi-Stakeholder Working Group on Data Governance. Participants said discussions had intensified in recent months following initial procedural delays. However, deep divergences remain. One fault line concerns whether data governance should focus primarily on individual privacy protections or incorporate broader societal and collective rights in data. Another centers on whether interoperability should be treated as an intrinsic public good or assessed in light of potential risks such as market concentration and data extractivism. A third area of contention involves cross-border data flows. Some frame free data flows with trust as essential for integration into global innovation markets. Others argue that asymmetries in infrastructure and bargaining power require preserving regulatory autonomy for developing countries, including the ability to pursue data sovereignty policies aligned with national development priorities.

Speakers emphasized that capacity building has emerged as an area of growing convergence. Governments across regions have called for technical, institutional, and policy support to develop interoperable systems and strengthen domestic data governance frameworks. While the final report will be brief, members indicated it would reflect divergent views without privileging majority positions. Discussions continue over whether the outcome should include recommendations or remain descriptive.

The working group is scheduled to meet again in Geneva in March, as it moves toward finalizing a document that proponents hope will guide international data governance debates in the years ahead.

Menaces hybrides: Comment améliorer la résilience face à ce phénomène?

The UN Institute for Disarmament Research (UNIDIR), in partnership with the Organisation internationale de la Francophonie (OIF), held an event to explore the phenomenon of hybrid threats. Experts recommended strengthening multilateral governance with harmonised norms for space and online platforms, building societal resilience through information and enforcement, and protecting critical infrastructure via cybersecurity and operational safeguards. Supporting less-equipped states with technical, regulatory, and risk-management tools is essential, alongside strategic signalling to make the consequences of unacceptable actions clear. Across all these measures, the guiding principle is integration—space, information, and cyber domains must be managed together to maintain global stability and resilience.

Launch of the World Intellectual Property Report 2026: Technology on the Move

The World Intellectual Property Organization (WIPO) launched the 2026 edition of its World Intellectual Property Report, entitled ‘Technology on the Move’, on 17 February (Tuesday) in Geneva and online. The report analyses how technologies spread globally and the implications for economic development. It reveals a dramatic acceleration in global technology diffusion: older technologies like the telegraph and automobile took decades to diffuse, whereas contemporary digital innovations, such as generative AI, reach users worldwide within days thanks to mature global digital infrastructure. Adoption gaps between advanced and developing economies have narrowed for recent technologies, and usage intensity differences are diminishing, especially for digital technologies. However, significant disparities remain, notably in Africa, where infrastructure and access gaps persist. Innovation leadership remains concentrated in a handful of economies, including the USA, Western Europe, Japan and China. Successful diffusion depends on four key factors—technology characteristics, information flow, absorptive capacity, and public policy and IP frameworks. The report stresses that deliberate policy and investment are essential to translate rapid diffusion into inclusive economic development and growth.

AI Summit in Geneva: Ten ways Switzerland can contribute to AI and humanity

In 2027, Geneva will host the next global AI Summit, arriving at a moment when governments, businesses, and communities worldwide are deep into AI-driven transformation.

Previous hosts brought their own distinctiveness: from Bletchley Park’s focus on existential risk to Seoul’s innovation-security balance, Paris’s economic and societal lens, and New Delhi’s emphasis on development and inclusion. Switzerland now has an opportunity to shape the next phase of AI governance and ensure that the 2027 AI Summit is more than just an event.

We suggest ten signposts on the Road to the 2027 Geneva AI Summit, supported by DiploAI research, training, and policy monitoring via the Digital Watch Observatory.

Innovation. AI is fundamentally about innovation, both technological and, increasingly, societal. The next wave of innovation will involve activating the knowledge of citizens and institutions through data labelling, embedding reinforcement learning into pedagogical practices, and developing knowledge graphs. Switzerland has long ranked among the world’s top innovators, favouring grounded, low-hype developments that address real needs and unexplored niches.

Governance. Existing international governance frameworks are likely to shape AI policy in Geneva, given the city’s concentration of organisations spanning trade, health, telecommunications, labour, and security. The new International Scientific Panel on AI can draw lessons from the Geneva-based IPCC’s experience at the science–diplomacy interface. Switzerland’s bottom-up policymaking model supports citizen inclusion in AI debates, while its cautious, gap-based regulatory tradition aligns with emerging calls for pragmatic, proportionate AI governance.

Subsidiarity. The principle of subsidiarity, central to Swiss societal organisation, holds that decision-making should occur as close as possible to the citizens and communities concerned. Applied to AI, this approach would counter the concentration of power in a few major platforms by rooting AI development in the local communities where knowledge is created through everyday interactions.

EspriTech. AI is prompting renewed reflection on fundamental questions of humanity, free will, and ethics, leading societies to revisit their cultural, religious, and philosophical foundations. Drawing on EspriTech and the intellectual legacy of Geneva’s thinkers, these lessons can help fine-tune debate on AI and humanity.

Trust. Switzerland, as a country with high trust capital, can foster a ‘trust but verify’ approach ahead of the 2027 Summit. Trust can be rebuilt through a fully informed and realistic discussion of AI risks, which have gradually recalibrated from 2023’s focus on existential risks (survival of humanity) to the current primacy of existing risks (education, disinformation, jobs) and growing concerns of exclusion risks (monopolisation of AI by a few actors).

Apprenticeship. The AI apprenticeship model, inspired by the Swiss tradition of learning by doing with mentorship, is emerging as an effective way to train in AI. Ahead of the 2027 Geneva AI Impact Summit, it can strengthen the AI knowledge and capacities of diplomats, civil society, and local communities.

Humanity. The 2027 AI Summit needs to give concrete meaning to the call for AI to serve humanity’s core interests. Switzerland has been ‘walking the talk’ of human-centred society in politics, education, social care, and the economy. This centuries-long experience can help fine-tune the critical connections between AI and human civilisation by both sharing some lessons learned and experimenting with new approaches and practices for the AI era.

Institutions. Institutions are an important carrier of societal memory and knowledge. AI should be considered a creative change agent that can, among other things, preserve institutional memory and strengthen the capacity to respond to societal needs.

Mutilateralism. The AI Summit can help clarify the purpose of international organisations in the AI era. AI can, for example, help foster a new level of legitimacy for international processes by ensuring that contributions to public consultations are properly traced and reflected in policy documents.

Sovereignty. As geopolitical tensions rise, questions of AI and digital sovereignty are becoming more relevant, highlighting the need for agency, self-determination, and responsible management of the knowledge that drives AI rather than isolation. Ahead of the 2027 Summit, Swiss experience can support more informed discussions on the technical, legal, and knowledge dimensions of AI sovereignty within an interdependent framework.

Read the suggestions in full at our dedicated web page.

BLOG featured image 2026 23 AI Summit in Geneva

www.diplomacy.edu

AI Summit in Geneva: Ten ways Switzerland can contribute to AI and humanity – Diplo

In 2027, Geneva will host the AI Summit.

Weekly #251 AI Summit in Geneva: Ten ways Switzerland can contribute to AI and humanity

20 – 27 February 2026

HIGHLIGHT OF THE WEEK

AI Summit in Geneva: Ten ways Switzerland can contribute to AI and humanity