Weekly #263 Russia–China Summit signals deepening cooperation across AI, industry, and digital economy

15 – 22 May 2026

HIGHLIGHT OF THE WEEK

Russia–China Summit signals deepening cooperation across AI, industry, and digital economy

Two high-level summits in as many weeks saw China host first the President of the United States and then the President of the Russian Federation. While the internet has done its thing, comparing Trump’s and Putin’s visits to China right down to the choreography of the children, we’ll focus our analysis on the tech topics that were discussed between Xi and Putin, with only a brief look at how the two sets of talks compare.

AI: Pushing a joint agenda. Putin and Xi emphasised that AI should be developed and deployed in ways that serve universal development rather than narrow national interests. At the same time, they explicitly criticised the use of AI by individual states as a geopolitical instrument to preserve or strengthen global dominance.

The leaders also highlighted growing cooperation on the military dimension of AI. Russia and China intend to deepen collaboration on the development and application of AI in military contexts, both through bilateral channels and within multilateral processes, including discussions under the Group of Governmental Experts on lethal autonomous weapons systems (LAWS).

The two sides committed to coordinating their positions on AI-related scientific and policy questions within international organisations. This indicated an effort to align their approaches more systematically in global standard-setting and norm-shaping forums. It’s not unusual for China and Russia to take similar positions in negotiations over digital matters, but this was a quite explicit geopolitical signal.

Person, Box, Face, Head, Cardboard, Carton, Package, Package Delivery

Beyond AI: Building parallel infrastructures. China and Russia will merge their GLONASS and Beidou satellite navigation systems into a complementary global service, cooperate in the coordination and use of radio frequencies and satellite orbital slots, and expand cooperation on satellite internet and the internet of things.

Open source cooperation also emerged as a strategic priority. The two countries plan to explore a bilateral software cooperation mechanism and to promote open-source technologies in key industries. That aligns with broader efforts in both countries to localise software ecosystems and reduce reliance on US tech platforms.

Cybersecurity. The parties agreed to strengthen cooperation in combating cybercrime, welcoming the UN General Assembly’s adoption of the UN Convention against Cybercrime and committing to its rapid entry into force and implementation, including work on an additional protocol to expand its scope and enhance international cooperation.

They pledged to deepen strategic cooperation on information security and coordinated responses to ICT threats. They also intend to exchange experiences in legislative regulation of the internet.

They emphasised the central role of the UN in responding to threats in the information space, after which the countries expressed support for the work of the Global Mechanism. This is also not surprising, considering that the Global Mechanism has arisen from the OEWG, which itself was a Russian initiative. The countries also expressed support for the development of broader international legal instruments covering issues such as data security and supply chain resilience. This aligns with China’s longtime push for establishing the Global Initiative on Data Security.

Digital economy. The parties agreed to expand cooperation on the digital economy and cross-border e-commerce, alongside broader collaboration in industries such as automotive, aviation, and mineral extraction. They pledged to strengthen coordination on intellectual property protection, consumer protection in online and cross-border services,

They also agreed to deepen e-commerce cooperation through regional frameworks, including the Eurasian Economic Union (EAEU), the Belt and Road Initiative, the Shanghai Cooperation Organisation (SCO), APEC, and the Enhanced Tumangan Initiative. Cooperation will focus on digitalisation, logistics, barrier-free trade, connectivity, and innovation, with the broader aim of boosting regional economic integration, trade, and employment.

This part is not surprising, considering that China and Russia have repeatedly framed e-commerce, the digital economy, and ICT cooperation as strategic pillars of their partnership since at least the mid-2010s in the SCO and Belt and Road contexts.

Comparing the two summits. There was no joint communique after the Xi–Trump meeting, while the Xi–Putin summit produced a detailed and lengthy one. That alone is quite telling. Xi and Putin also signed over 20 agreements covering energy, trade, science and technology, and infrastructure, while it is still unclear how many agreements Xi and Trump signed.

In the US–China context, recent comments made by US Treasury Secretary Bessent during Trump’s visit to China, noting that ‘The reason we are able to have wholesome discussions with the Chinese on AI is because we are in the lead,’ highlight continued tech competition between the two countries. President Trump also just postponed signing the AI and cybersecurity order, saying he did not want new rules to slow US leadership in AI or weaken its competitive edge over China, which reinforces the competitive framing.

This is in direct opposition to cooperation with Russia and to future coordination on AI-related scientific and policy questions within international organisations.

After the Trump-Xi summit, there was initially no indication that rare earths had been discussed. However, the White House later stated that China will address US concerns regarding supply chain shortages related to rare earths and other critical minerals, as well as concerns over restrictions on the sale of rare earth production and processing equipment and technologies. In response to questions about this statement, China’s Ministry of Commerce said that both sides had discussed the issue and would work to resolve each other’s reasonable and lawful concerns.

China’s engagement with Russia took a more cooperative and forward-looking form. The two sides agreed to actively promote cooperation in joint mineral extraction and in developing green standards. Russia holds some of the world’s largest rare earth reserves—ranked fifth globally—but remains constrained by limited capacity in extraction, refining, and processing, an area where it stands to benefit from Chinese technological and industrial expertise.

Overall, the difference in both outcomes and issue salience indicates that the Russia–China relationship appears more comfortable addressing sensitive domains such as AI cooperation and military applications, whereas US–China engagement remains more constrained and competitive in tone and scope.

IN OTHER NEWS LAST WEEK

AI governance

Vatican. Pope Leo XIV’s first encyclical, Magnifica humanitas, focusing on the protection of the human person in the age of AI, will be released on 25 May. The document, whose title roughly translates as ‘Magnificent humanity,’ addresses the relationship among AI, human dignity, and Catholic social teaching. The Vatican has also established an Inter-Dicasterial Commission on Artificial Intelligence, approved by Pope Leo XIV, to coordinate work on the implications of rapidly advancing AI technologies.

The EU. The European Commission has published a draft guidance related to high-risk AI systems under the EU AI Act. The draft guidance is intended to help providers and deployers determine whether AI systems fall within the Act’s high-risk category. The document includes examples illustrating how classification criteria may apply in different situations. The document has been made available through the AI Act Single Information Platform. Stakeholders, including businesses, public authorities, researchers, civil society organisations, and citizens, may submit comments until 23 June 2026.

The USA. A federal appeals court in Washington appeared divided over Anthropic’s challenge to the Pentagon’s decision to designate the company as a national security supply chain risk. During oral arguments, judges on the D.C. Circuit questioned both the factual basis and the scope of the Pentagon’s authority, with one judge describing the designation as a ‘spectacular overreach,’ while another pressed whether courts should defer to the Defence Secretary’s judgement on national security risk assessments.

Meanwhile, Anthropic has revised disclosure rules for its Project Glasswing cybersecurity programme, allowing participating partners using its Mythos model to share vulnerability findings, tools and mitigation insights more broadly with regulators, industry bodies and, where appropriate, the public under responsible-disclosure norms. The shift marks a departure from earlier confidentiality requirements that kept findings within the programme

A federal jury in California ruled in favour of OpenAI, Sam Altman and Greg Brockman in Elon Musk’s lawsuit alleging that the company abandoned its original non-profit mission after transitioning toward a for-profit structure. The court accepted OpenAI’s argument that Musk had long been aware of discussions around restructuring, concluding that the claims were filed outside the applicable legal time limits and dismissing the case following an advisory jury verdict.

Malta. Malta has launched AI for All, a free national AI literacy programme for Maltese citizens and residents with an active eID account. The initiative is led by the Malta Digital Innovation Authority, with the educational programme developed by the University of Malta. The programme is intended to help participants understand AI capabilities, limitations, and responsible use in everyday life, education, and work.

The UK. Ofcom has announced a recommendation intended to strengthen protections against illegal intimate image abuse online. The UK regulator said it is updating its Illegal Content Codes to recommend that certain online platforms use automated detection technologies to identify such images, including AI-generated explicit deepfakes and non-consensual image sharing. The recommendation is expected to enter into force in autumn 2026, subject to parliamentary approval.

Canada-Spain. Canada and Spain have signed a memorandum of understanding to strengthen cooperation in AI development, adoption, and digital innovation. The memorandum establishes a framework for cooperation on AI technologies, computing capacity, and collaboration between governments, industry, and other stakeholders. According to both governments, the partnership aims to support innovation, investment, and economic growth linked to AI technologies.

Platforms continue to face pressure over youth safety online

In the USA, bellwether trials continue. YouTube, Snap, TikTok, and Meta have agreed to settle a landmark lawsuit brought by Kentucky’s Breathitt County School District, which accused them of designing algorithmic systems and features—such as infinite scrolling and engagement-driven recommendation loops—that foster compulsive use among young users and place additional financial and educational burdens on schools. Terms of the settlement have not been disclosed. The case was scheduled to go to trial on 12 June, which will now not proceed.

In parallel, the UK’s Ofcom has reported that almost all 8-17-year-olds in the UK are online (99%). reinforcing its push for stronger platform accountability. It wants child safety to be embedded into product design from the outset, with regulators informed in advance of new features to assess potential risks. However, the regulator remains concerned that some platforms are not doing enough to make feeds safer and is sceptical that age-13 minimum policies effectively prevent underage access.
Malaysia has also announced new rules requiring stronger safeguards for young users on online platforms, including stricter age-related controls and enhanced content governance obligations for service providers. The measures are part of a broader policy effort to reduce exposure of minors to harmful online content and to formalise platform responsibility in youth protection.

Australia introduces a national framework for digital health standards

Australia has introduced a National Framework for Digital Health Standards to improve interoperability and consistency across healthcare systems. The framework is intended to support integration of digital tools and health records across healthcare settings.

The framework addresses fragmentation caused by independently developed digital health standards. It also provides guidance intended to support coordination between government agencies, healthcare providers, and industry participants.

The framework also supports the use of internationally recognised clinical terminology standards and related training initiatives.

OECD review highlights growth and regulatory challenges in ASEAN digital trade

The OECD has published a Digital Trade Review of the Association of Southeast Asian Nations, examining regional growth in digital trade and related regulatory challenges.

The OECD said ASEAN benefits from trade openness, increasing digital adoption, and evolving regional policy initiatives. The report noted that uneven participation and fragmented domestic regulations may limit further digital trade integration across the region.

The review identified barriers, including restrictions affecting cross-border data flows, telecommunications, digital services, and trade facilitation systems. The OECD highlighted the importance of regulatory alignment and progress towards paperless trade systems.

The report also discussed opportunities related to AI adoption, including reforms linked to tariffs, data flows, and digital services regulation. These findings underline the importance of coordinated reforms to strengthen ASEAN’s role in the global digital economy.

View all updates

LAST WEEK IN GENEVA

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

79th World Health Assembly

The 79th World Health Assembly (WHA79), which took place in Geneva from 18–23 May 2026, brought together WHO Member States, international organisations, civil society, and health stakeholders to discuss global health priorities and governance. Alongside the formal Assembly proceedings, a wide range of side events focused on digital health and AI, covering topics such as trusted health data, interoperability, responsible AI governance, and digital transformation of health systems. Discussions will examine how AI and digital technologies can support universal health coverage, strengthen evidence-based policymaking, improve diagnostics and ageing care, and enhance cross-border cooperation, while also addressing issues of ethics, transparency, data governance, and equity in healthcare systems.

UN human rights experts raise concerns over the large-scale sexual exploitation of women and girls online

The UN Special Rapporteur on violence against women and girls, its causes and consequences and the UN Special Rapporteur on the right to privacy issued a statement raising concerns over what they describe as large-scale sexual exploitation of women and girls facilitated and monetised through Pornhub and its parent company, Aylo Holdings, as well as by payment networks and search engines that enable distribution and monetisation of content.

The rapporteurs focused on Pornhub and its parent company Aylo Holdings, while also referencing broader concerns involving user-generated pornography platforms, payment networks, and search and technology companies linked to online distribution systems.

The experts said businesses involved in digital content ecosystems should not avoid responsibility where their services contribute to human rights violations. They called for stronger safeguards, including mandatory third-party age and consent verification systems for user-generated pornography platforms.

LOOKING AHEAD

EuroDIG 2026

The next EuroDIG will take place on 26-27 May 2026 at the Charlemagne Building of the European Commission in Brussels, under the theme ‘European Voices for the Future of the Internet – Celebrating 20 Years of .eu and the Beginning of a New Internet Governance Era’. It will be hosted by EURid – the registry for the .eu domain name and supported by the European Commission, a longstanding Institutional Partner of EuroDIG. The event will mark 20 years of .eu, celebrating two decades of the trusted digital identity. Diplo and GIP will provide reports from a selected number of sessions from the event—make sure to bookmark the dedicated web page.

READING CORNER

BLOG featured image 2026 54 Put the internet on the World Heritage List instead

Put the internet on the World Heritage List instead – Diplo

What if the world lost its UNESCO World Heritage sites? We would mourn them, certainly. But how much more would our lives change if we lost language, the scientific method, double-entry bookkeeping, or the internet? Aldo Matteucci examines.

Lesson from Singapore: Why diplomats should understand AI before negotiating its governance – Diplo

Not everyone needs to become a programmer. But everyone involved in AI governance should understand the technology well enough to ask pertinent questions, challenge assumptions, and identify realistic policy options. The Singaporean minister’s example can help overcome the current gap between high interest in AI governance and low understanding of new technology, Jovan Kurbalija argues.

OPPORTUNITY

Opportunity: Become a Knowledge Fellow

Diplo is pleased to launch a new call for applications for Digital Watch Knowledge Fellows (2026), the team of collaborators behind the Digital Watch Observatory (DW). Knowledge Fellows (KF) are central to the observatory’s ability to provide comprehensive, accurate, and up-to-date coverage of specific areas of digital governance. More details on what we are looking for and what we offer in return are available here. Interested applicants are invited to apply by 31 May 2026.

Art, Modern Art, Graphics, Cutlery, Fork, Text

Lettre d’information du Digital Watch – Numéro 109 – Mensuelle avril 2026

Rétrospective de avril 2026

Les déclarations utopiques ou extravagantes sur l’avenir sont monnaie courante dans le discours du secteur technologique. Mais l’accélération fulgurante de l’IA générative semble avoir raccourci le délai de concrétisation de bon nombre de ces affirmations. Le mois d’avril a vu déferler une nouvelle vague de prédictions très médiatisées. Si certains pourraient être tentés de les rejeter comme du simple battage médiatique, il existe une raison de les prendre au sérieux : le danger insidieux de concevoir l’avenir sans une participation significative du public.

L’Afrique du Sud a dévoilé son premier projet de politique nationale en matière d’IA et a rapidement été contrainte de le retirer après que des examinateurs y ont découvert une faille critique : il était truffé de fausses sources et de citations imaginaires, probablement générées par l’IA. Cet incident illustre bien un problème plus général lié aux lois générées par l’IA. Dans ce numéro de la newsletter, nous examinons comment empêcher que de fausses lois ne régissent la vie réelle.

Début avril, Anthropic a annoncé Claude Mythos Preview, son modèle d’IA le plus performant à ce jour, tout en prenant la décision explicite de ne pas le rendre accessible au public. Nous nous penchons sur les capacités du modèle, la raison qui justifie la restriction de son accès, ainsi que les questions de gouvernance que ce modèle soulève.

Nous invitons les lecteurs intéressés à rejoindre notre équipe de Knowledge Fellows. Les Knowledge Fellows jouent un rôle central dans la capacité de l’observatoire à fournir une couverture complète, précise et actualisée de domaines spécifiques de la gouvernance numérique. Vous trouverez dans la newsletter plus de détails sur ce que nous recherchons et ce que nous offrons en contrepartie.

En bonus : les principales évolutions en matière de politique numérique au mois d’avril et un compte rendu de Genève.

Coup d’oeil : les événements qui ont fait la une en avril

TECHOLOGIES

L’UE et les États-Unis ont mis en place un cadre coordonné visant à renforcer la résilience des chaînes d’approvisionnement en minéraux critiques, combinant un protocole d’accord stratégique (MoU) et un plan d’action. Ce partenariat a pour objectifs de garantir des chaînes d’approvisionnement diversifiées et durables grâce au développement de projets communs au sein de l’UE, des États-Unis et dans des pays tiers, avec le soutien d’outils d’investissement coordonnés, des mécanismes de réduction des risques et des liens commerciaux renforcés.

Le Canada et la Finlande ont défini un nouveau programme de coopération en matière de technologies souveraines et d’IA, plaçant les capacités numériques avancées au cœur de la résilience économique, de la sécurité et de l’autonomie stratégique dans un environnement mondial contesté. Annoncé à l’issue de discussions à Ottawa, ce programme couvre l’adoption de l’IA au sein du gouvernement et de l’industrie, le calcul haute performance, les télécommunications, les usines géantes d’IA (y compris le soutien à la gigafactory d’IA de Nokia), la recherche quantique, les minéraux critiques et les chaînes d’approvisionnement fiables. Les deux pays prévoient d’approfondir leur coordination en matière d’infrastructures souveraines d’IA, de réduire les dépendances technologiques, de soutenir les petites et moyennes entreprises et d’élargir les opportunités dans le domaine des télécommunications grâce à des initiatives telles que la Coalition mondiale sur les télécommunications.

Le Canada renforce son soutien à l’écosystème de la recherche quantique grâce à un nouveau financement annoncé par le Conseil de recherches en sciences naturelles et en génie du Canada, visant à consolider les capacités scientifiques du pays, son potentiel d’innovation et son leadership à long terme dans un domaine d’importance stratégique. Cette initiative soutiendra les chercheurs, les projets et la collaboration interinstitutionnelle, favorisant à la fois la science fondamentale et le développement appliqué tout en contribuant à concrétiser les avancées de la recherche quantique en progrès technologiques concrets.

Le gouvernement britannique a identifié six technologies de pointe – l’IA, la cybersécurité, la connectivité avancée, la biologie d’ingénierie, les technologies quantiques et les semi-conducteurs – comme les piliers de sa Stratégie industrielle moderne 2025 et de son Plan sectoriel pour le numérique et les technologies, visant à renforcer les capacités numériques, la croissance économique, la résilience nationale et la compétitivité à long terme. Ce programme donne la priorité aux investissements dans les télécommunications de nouvelle génération, notamment la 5G et la future 6G, ainsi qu’à l’augmentation de la capacité de calcul, aux infrastructures de supercalcul et au développement de la main-d’œuvre afin de renforcer la position du Royaume-Uni en tant que pôle de premier plan de l’UE en matière d’IA.

Des chercheurs australiens ont utilisé un modèle d’IA basé sur Wikipédia pour identifier 100 technologies émergentes qui devraient prendre de l’ampleur d’ici 2026, offrant ainsi une alternative fondée sur les données aux méthodes de prévision traditionnelles, souvent influencées par l’avis d’experts. S’appuyant sur des milliers d’entrées Wikipédia, l’analyse a répertorié plus de 23 000 technologies pour établir la liste « Momentum 100 », dominée par l’apprentissage par renforcement et suivie par la blockchain, l’impression 3D, la robotique souple, la réalité augmentée et d’autres secteurs en plein essor.

Infrastructure

Les fournisseurs de technologies européens Cubbit, SUSE, Elemento et StorPool Storage ont lancé un pack de reprise après sinistre commun afin d’aider les organisations à maintenir l’accès aux données et la continuité opérationnelle en cas de perturbations liées à des dépendances technologiques externes. Présentée lors du Sommet européen des données à Berlin, cette solution combine stockage, calcul, coordination, mise en réseau, gestion des identités, observabilité et gestion au sein d’une seule plateforme logicielle cloud déployable, conçue pour réduire la fragmentation et simplifier la planification de la reprise après sinistre. En permettant le transfert des charges de travail critiques vers une infrastructure basée en Europe avec un minimum de perturbations, cette initiative vise à répondre aux besoins concrets en matière de reprise après sinistre tout en soutenant les efforts plus larges visant à réduire la dépendance vis-à-vis des fournisseurs de cloud non européens.

Un nouveau rapport, s’appuyant sur les recherches menées par le Future of Technology Institute, basé à Bruxelles, met en garde sur le fait que la plupart des agences de défense de l’UE restent fortement dépendantes des fournisseurs de cloud et de technologies américains, ce qui soulève des inquiétudes quant à l’exposition à un scénario potentiel de « kill switch » (coupe-circuit numérique) dans lequel des services critiques pourraient être restreints ou désactivés en cas de tensions politiques ou stratégiques. Les données sur les marchés publics examinées par l’institut suggèrent que 23 des 28 pays de l’UE et du Royaume-Uni s’appuient sur des entreprises américaines, soit directement, soit par l’intermédiaire de fournisseurs de l’UE, en utilisant une infrastructure cloud américaine, 16 pays étant classés à haut risque, dont l’Allemagne, la Finlande, la Pologne, le Danemark, l’Estonie et le Royaume-Uni. Google Cloud, Microsoft et Oracle sont décrits comme des fournisseurs dominants dans les systèmes de défense sensibles, tandis que l’Autriche est présentée comme un cas à moindre risque en raison de son recours apparent à des alternatives souveraines.

Panthalassa a levé 140 millions de dollars dans le cadre d’un financement de série B, mené par Peter Thiel, pour développer des systèmes offshore exploitant l’énergie des vagues océaniques afin d’alimenter le calcul IA alors que la demande en capacité des centres de données s’accélère. L’entreprise prévoit de construire des nœuds alimentés par l’énergie des vagues qui produiront de l’électricité en mer, exécuteront des calculs d’IA à bord et transmettront les données via des satellites en orbite basse, offrant ainsi une solution aux contraintes croissantes des centres de données terrestres en matière d’alimentation électrique, de refroidissement et d’infrastructure.

SÉCURITÉ

En associant la découverte basée sur l’IA à une coordination sectorielle, 100 millions de dollars de crédits d’utilisation et à un financement dédié à la sécurité open source, le projet Glasswing rassemble des acteurs majeurs des domaines de la technologie, de la cybersécurité, de la finance et de l’open source, notamment AWS, Apple, Google, Microsoft, NVIDIA, Cisco, CrowdStrike, JPMorgan Chase et la Linux Foundation, dans le cadre d’une initiative conjointe visant à utiliser l’IA de pointe pour défendre les infrastructures logicielles critiques. Dirigée par le modèle Claude Mythos Preview d’Anthropic, cette initiative vise à détecter des vulnérabilités complexes à grande échelle. Ses premiers résultats ont permis de mettre à jour des milliers de failles jusque-là inconnues dans les systèmes d’exploitation, les navigateurs et les infrastructures numériques de base, dont certaines étaient restées dissimulées pendant des décennies.

Un avis conjoint de la CISA met en garde contre le fait que des cyberacteurs affiliés à l’Iran ciblent des automates programmables connectés à Internet au sein des infrastructures critiques américaines, notamment les appareils Rockwell Automation et Allen-Bradley CompactLogix et Micro850 utilisés dans les systèmes gouvernementaux, d’approvisionnement en eau, énergétiques et industriels. Active depuis au moins mars 2026, cette campagne a perturbé les fonctions des automates programmables, manipulé des fichiers de projet et altéré les affichages des IHM et des systèmes SCADA, causant des dommages opérationnels et financiers.

Le Canada a lancé le niveau 1 du Programme canadien de certification en cybersécurité, établissant ainsi des exigences de base en matière de cybersécurité pour les fournisseurs travaillant sur des contrats de défense, alors que les cybermenaces visent de plus en plus les sous-traitants, les données sensibles et les chaînes d’approvisionnement critiques. La mise en œuvre progressive débutera à l’été 2026, la certification étant requise dès l’attribution du contrat, ce qui laissera à l’industrie le temps de s’adapter tout en renforçant la confiance dans les marchés publics et la préparation opérationnelle.

L’évaluation 2026 d’Europol des menaces liées à la cybercriminalité organisée sur Internet met en garde contre le fait que le paysage de la cybercriminalité dans l’UE devient plus complexe, plus industrialisé et plus difficile à contrer, les réseaux criminels exploitant le chiffrement, les serveurs proxy, la fragmentation des espaces en ligne et les outils basés sur l’IA. Le rapport identifie les facilitateurs de la cybercriminalité, la fraude en ligne, les cyberattaques et l’exploitation sexuelle des enfants en ligne comme des domaines de préoccupation majeurs, l’IA rendant les escroqueries, la tromperie et les abus plus évolutifs et plus convaincants.

La Norvège a annoncé son intention d’interdire l’utilisation des réseaux sociaux aux enfants de moins de 16 ans, en confiant la responsabilité de la vérification de l’âge aux entreprises technologiques.

La Grèce s’apprête à renforcer les restrictions sur l’utilisation des réseaux sociaux par les mineurs, avec une législation attendue dans le courant de l’année qui introduirait une interdiction pour les enfants de moins de 15 ans. La mesure devrait entrer en vigueur le 1er janvier 2027 et vise à établir un cadre qui modifie le fonctionnement des plateformes. Les plateformes seraient tenues de mettre en place des mécanismes rigoureux de vérification de l’âge, y compris la revérification des comptes existants, sous la supervision d’autorités nationales de régulation telles que la Commission hellénique des télécommunications et des postes (EETT).

Le président français Emmanuel Macron convoque les dirigeants de l’UE, dont le Premier ministre espagnol Pedro Sanchez et des représentants de l’Italie, des Pays-Bas et de l’Irlande, afin d’harmoniser les approches nationales pour restreindre l’accès des mineurs aux réseaux sociaux et de faire pression pour une action plus rapide au niveau européen.

Le projet de loi britannique sur le bien-être des enfants et les écoles (Children’s Wellbeing and Schools Bill) vise à étendre les pouvoirs des ministres pour influencer la manière dont les services en ligne protègent les enfants, notamment en restreignant l’accès aux plateformes, fonctionnalités ou fonctions à risque et en ciblant des éléments de conception tels que les paramètres de contact, la communication en direct, la visibilité de la localisation et le temps passé en ligne. Le projet conférerait également à l’Ofcom un rôle consultatif plus important, introduirait un délai de six mois pour l’adoption de réglementations ou la présentation d’un rapport d’étape, et donnerait aux ministres de nouveaux pouvoirs en matière de consentement relatif aux données des enfants, de vérification de l’âge et d’application de la loi. Ce dispositif réglementaire reste pour l’instant en suspens, le Parlement négociant toujours des mesures clés et aucune loi définitive n’étant encore en vigueur.

La Commission européenne a développé une application standardisée de vérification de l’âge destinée à fonctionner dans tous les États membres. Cette application permet aux utilisateurs de confirmer qu’ils remplissent les conditions d’âge requises pour accéder aux plateformes de réseaux sociaux en fournissant leur numéro de passeport ou de carte d’identité. Elle est conçue pour s’intégrer dans les portefeuilles numériques nationaux ou fonctionner comme une application autonome, dans le cadre d’un dispositif européen coordonné visant à garantir l’interopérabilité et à éviter la fragmentation des systèmes nationaux. L’application est open source et disponible pour une mise en œuvre tant publique que privée, mais elle est soumise à des exigences techniques et de confidentialité communes. La Commission prévoit de mettre en place un mécanisme de coordination au niveau de l’UE pour superviser le déploiement, l’accréditation et l’utilisabilité transfrontalière. Le déploiement a fait l’objet d’un examen minutieux de la part des chercheurs en sécurité. Parmi les faiblesses signalées figurent des données d’authentification stockées localement qui peuvent être réinitialisées ou modifiées, permettant ainsi aux utilisateurs de contourner les protections par code PIN, de désactiver les contrôles biométriques et de réinitialiser les mécanismes de limitation de débit en modifiant les fichiers de configuration. Cela permet en effet la réutilisation de données d’identité vérifiées sous des contrôles d’accès modifiés. Ces critiques ont suscité des inquiétudes plus générales parmi les développeurs concernant l’architecture de l’application, notamment sur les raisons pour lesquelles des fonctionnalités matérielles sécurisées n’ont pas été utilisées, et sur la nécessité logique d’éléments tels que l’expiration des informations d’âge.

La Commission européenne a également pris récemment des mesures préliminaires à l’encontre de Meta, estimant que Facebook et Instagram n’ont pas empêché efficacement les utilisateurs de moins de 13 ans d’accéder à leurs services, en grande partie parce que les contrôles d’âge peuvent être contournés à l’aide de fausses dates de naissance et de systèmes de vérification peu fiables.

En Australie, la campagne en faveur de la sécurité des enfants s’étend des réseaux sociaux aux jeux vidéo, alors que les régulateurs intensifient leur surveillance de la manière dont les plateformes protègent les mineurs contre les dangers. Le 21 avril, le commissaire à la sécurité en ligne a émis des avis de transparence juridiquement contraignants à l’encontre de Roblox, Minecraft, Fortnite et Steam, exigeant des détails sur la manière dont ces plateformes gèrent les risques, notamment l’exploitation sexuelle des enfants, le cyberharcèlement, les discours de haine et les contenus extrémistes sur des services largement utilisés par les enfants.

Le Bureau du commissaire à l’information du Royaume-Uni a lancé une campagne pour aider les parents et les tuteurs à parler de la confidentialité en ligne avec les enfants en âge d’aller à l’école primaire, après qu’une étude a révélé que de nombreux enfants partagent des informations personnelles en ligne, tandis que les familles ne savent souvent pas comment réagir. L’ICO indique que 24 % des enfants ont partagé leur vrai nom ou leur adresse en ligne, 22 % ont divulgué des informations telles que des détails sur leur santé à des outils d’IA, et 21 % des parents n’ont jamais discuté de la confidentialité en ligne avec eux.

ÉCONOMIE

La Commission européenne a adressé à Meta un acte d’accusation complémentaire (intitulé « communication des griefs complémentaire »), soulignant ses préoccupations concernant d’éventuelles restrictions à l’accès des assistants IA tiers à WhatsApp. Auparavant, Meta avait décidé de rétablir l’accès à WhatsApp pour les assistants IA tiers moyennant des frais. Cependant, la Commission a conclu à titre préliminaire que ces mesures restaient anticoncurrentielles et a désormais pris des mesures provisoires pour empêcher que ces changements de politique ne causent un préjudice grave au marché. Ces mesures provisoires resteront en vigueur jusqu’à ce que la Commission ait mené à bien son enquête et rende une décision finale sur le comportement de Meta.

La CNUCED indique que le commerce mondial a augmenté de 2 500 milliards de dollars en 2025 pour atteindre 35 000 milliards de dollars, reflétant une expansion continue des biens et des services, mais aussi un paysage économique plus fragile et inégal. La montée des tensions géopolitiques, la perturbation des routes maritimes, les conflits au Moyen-Orient et l’instabilité dans les principaux couloirs maritimes font grimper les coûts de l’énergie, des transports et des importations, exerçant une pression accrue sur les économies en développement disposant d’une marge de manœuvre budgétaire limitée. La croissance des services a ralenti, tandis qu’une grande partie de la récente augmentation des échanges commerciaux résulte davantage de la hausse des prix que d’une augmentation des volumes. L’Asie de l’Est et l’Afrique restent des moteurs importants grâce au commerce Sud-Sud et à la réorganisation des chaînes d’approvisionnement, mais la fragmentation, le découplage entre les États-Unis et la Chine, l’inflation, l’endettement et le protectionnisme devraient peser sur les perspectives pour 2026.

L’Organisation internationale du travail met en garde contre le fait que les systèmes de protection sociale ne parviennent pas à suivre le rythme de l’évolution rapide des marchés du travail, influencés par le changement climatique, les bouleversements technologiques, les changements démographiques et l’évolution des modalités de travail. Son nouveau rapport met en évidence des lacunes majeures en matière de couverture, d’adéquation et de financement, laissant de nombreux travailleurs sans protection en cas de chômage, de maladie, de retraite ou de transition professionnelle.

La Russie s’apprête à ériger en infraction pénales les activités non autorisées à grande échelle liées aux cryptomonnaies, après qu’une commission législative gouvernementale a approuvé des amendements prévoyant des peines de prison pour l’organisation de la circulation de monnaies numériques sans l’autorisation de la banque centrale. L’article 171.7 proposé du Code pénal sanctionnerait les cas impliquant un préjudice important, des revenus illicites considérables ou des dommages causés à des particuliers, à des organisations ou à l’État d’une peine de 4 à 7 ans de prison. Cette mesure, qui devrait entrer en vigueur le 1er juillet 2027, marque un durcissement de la répression dans le secteur des actifs numériques en Russie.

La Commission européenne a mis à jour ses règles de concurrence en matière de transfert de technologies afin de mieux refléter l’innovation fondée sur les données, les marchés numériques et les pratiques modernes en matière de licences dans l’ensemble de l’UE. Le cadre révisé clarifie la manière dont les entreprises peuvent concéder des licences pour des brevets, des logiciels, du savoir-faire et des technologies liées aux données tout en respectant le droit de la concurrence, dans le but de protéger la collaboration et la sécurité juridique sans autoriser les accords qui restreignent l’accès au marché ou l’innovation. Une attention accrue est accordée aux écosystèmes numériques, aux technologies essentielles aux normes et aux accords de licence susceptibles d’influencer le contrôle des données, l’interopérabilité et la concurrence en aval.Le Canada a annoncé un financement de 23,8 millions de dollars canadiens pour le programme « Compétences numériques pour les jeunes », visant à aider les jeunes à acquérir une expérience pratique alors que l’IA, la cybersécurité, le big data, l’automatisation et la transformation numérique au sens large redessinent le marché du travail. Dirigée par la ministre de l’Industrie, Mélanie Joly, cette initiative d’une durée de deux ans financera des formations et des stages pour les diplômés de l’enseignement supérieur en les mettant en relation avec des employeurs des secteurs des technologies émergentes. Les bénéficiaires éligibles sont les entreprises, les organisations à but non lucratif, les institutions publiques, les organisations autochtones et les organismes provinciaux ou territoriaux, avec un accès flexible pour les participants du Yukon, des Territoires du Nord-Ouest et du Nunavut.

DROITS DE L’HOMME

Le Brésil a inauguré son premier Centre d’accès, de recherche et d’innovation en technologies d’assistance (Capta) à l’Institut Benjamin Constant de Rio de Janeiro. Géré par le ministère des Sciences, de la Technologie et de l’Innovation (MCTI) dans le cadre du Plan national pour les droits des personnes handicapées, ce centre a pour objectif de favoriser le développement, l’expérimentation et la diffusion de technologies d’assistance qui renforcent l’autonomie, l’inclusion et la qualité de vie des personnes handicapées. Ce lancement marque la création du premier d’une série de centres prévus à l’échelle nationale pour élargir l’accès à ces technologies.

L’UNESCO met en garde contre le fait que les élèves en situation de handicap continuent de se heurter à des obstacles majeurs dans le domaine de l’éducation, notamment des infrastructures inaccessibles, des technologies d’assistance limitées, une formation insuffisante des enseignants, la stigmatisation et des systèmes de données défaillants qui laissent de nombreux apprenants invisibles dans la planification des politiques. Ses conclusions montrent que l’exclusion commence souvent très tôt et est renforcée par la pauvreté, les inégalités entre les sexes, les déplacements de population et d’autres désavantages qui se cumulent, limitant ainsi l’accès à un apprentissage de qualité et aux opportunités futures. L’UNESCO exhorte les gouvernements à aller au-delà de mesures d’inclusion étroites en investissant dans des écoles accessibles, des programmes d’études inclusifs, des éducateurs formés, des données fiables et une participation significative des personnes en situation de handicap.

Les Philippines et les Bermudes ont signé un protocole d’accord visant à renforcer la coopération transfrontalière en matière de protection des données à caractère personnel, établissant un lien entre la Commission nationale de la vie privée des Philippines et le Bureau du commissaire à la protection de la vie privée des Bermudes. Cet accord permet le partage d’informations, l’assistance mutuelle dans le cadre d’enquêtes et une coordination plus étroite sur les cas de violation de données relevant de plusieurs juridictions. Au-delà de l’application de la loi, ce partenariat soutient la mise en place de mécanismes compatibles de protection des données, de cadres de certification, de flux de données fiables, de formations et d’échanges de connaissances sur les nouveaux défis en matière de protection de la vie privée.

JURIDIQUE

Une décision unanime de la Cour suprême des États-Unis a restreint les circonstances dans lesquelles un fournisseur d’accès à Internet (FAI) peut être tenu responsable de la violation des droits d’auteur par ses utilisateurs. S’exprimant au nom de la Cour, le juge Clarence Thomas a déclaré qu’un FAI n’est responsable que si son service a été conçu pour des activités illégales ou s’il a activement incité à la violation.

Les autorités françaises ont convoqué Elon Musk et l’ancienne directrice de X, Linda Yaccarino, pour qu’ils se présentent à des auditions volontaires dans le cadre d’une enquête pénale visant à déterminer si X a facilité la diffusion de contenus pédopornographiques, de deepfakes générés par l’IA, de contenus niant l’Holocauste et d’autres contenus préjudiciables ou illégaux. Cependant, M. Musk semble avoir refusé de se présenter. Le conflit s’est amplifié lorsque des informations ont révélé que le ministère américain de la Justice avait refusé d’apporter son aide à l’enquête française, arguant que l’affaire risquait d’empiéter sur la réglementation de la liberté d’expression et qu’elle viserait injustement une entreprise américaine. Les autorités françaises ont toutefois présenté cette affaire comme une mesure d’application légitime de la loi nationale.

Dans le cadre du litige fédéral multidistrict (MDL) en cours devant le district nord de Californie impliquant Meta, Google (YouTube), ByteDance (TikTok) et Snap Inc., le tribunal a rejeté les requêtes en irrecevabilité déposées par plusieurs établissements scolaires. Cette décision fait passer l’affaire de la phase de plaidoirie à celle des procédures de référence, au cours desquelles des affaires sélectionnées permettront de tester les principes fondamentaux en matière de responsabilité et de dommages-intérêts. L’argument principal des plaignants repose sur la conception des produits. Ils affirment que les plateformes ont été conçues pour maximiser l’engagement des mineurs malgré la connaissance interne des risques pour la santé mentale. Ils établissent un lien entre cette conception et l’augmentation signalée des cas d’anxiété, de dépression et de troubles du comportement en milieu scolaire. La chaîne de causalité est contestée, mais c’est la théorie centrale avancée. La chaîne de causalité est contestée, mais il s’agit là de la théorie centrale avancée. Le MDL est vaste, avec plus de 2 300 actions connexes réparties dans six États, ce qui en fait l’un des litiges les plus importants dans ce domaine. Le procès de référence prévu en juin devrait constituer le premier véritable test de ces allégations et influencera probablement à la fois la pression en faveur d’un règlement et l’orientation générale du MDL.

L’affaire Raine c. OpenAI est traitée comme une affaire distincte en Californie et ne fait partie d’aucun MDL. La plainte allègue que l’utilisation de ChatGPT par Adam Raine est passée d’une finalité académique à une dépendance émotionnelle, les révélations de plus en plus fréquentes concernant sa santé mentale ayant prétendument reçu des réponses qui ont renforcé sa dépendance plutôt que de l’orienter vers l’extérieur. Les plaignants soutiennent qu’il s’agissait là d’une conséquence prévisible d’une conception axée sur l’engagement. Ils font valoir des chefs d’accusation, notamment celui de mort par négligence, et demandent des mesures injonctives visant à renforcer les mesures de protection. Bien qu’aucune date de procès n’ait été officiellement fixée, l’affaire en est encore à ses débuts sur le plan procédural en Californie et pourrait être jugée fin 2026 ou en 2027, en fonction de l’évolution de la situation avant le procès.

SOCIOCULTUREL

La Commission européenne a lancé un programme de transformation numérique en Méditerranée destiné aux pays d’Afrique du Nord et du Moyen-Orient, marquant ainsi la première initiative numérique dans le cadre du Pacte pour la Méditerranée. Ce programme vise à soutenir une croissance inclusive et durable en améliorant l’accès aux services numériques, en alignant la réglementation des télécommunications sur les normes de l’UE et en renforçant les autorités réglementaires nationales.La cybersécurité constitue une priorité essentielle, avec un soutien au renforcement des cadres nationaux, des capacités institutionnelles et des réponses coordonnées aux menaces numériques.

Les premiers résultats du suivi mené par la Commission européenne au titre du Code de conduite révisé sur la lutte contre les discours de haine illicites en ligne+ montrent que les grandes plateformes progressent dans l’examen des discours de haine illicites signalés dans un délai de 24 heures, même si des lacunes subsistent en matière d’exactitude, de cohérence et de pratiques de signalement. Sur la base d’un suivi indépendant et des données fournies par les entreprises, l’évaluation a révélé que de nombreuses notifications avaient été traitées dans les délais convenus, mais qu’une part notable des cas avait fait l’objet de contestations ou avait été classée à tort. Lié au modèle de corégulation de la loi sur les services numériques, cet exercice sert de test pratique de la responsabilité, de la transparence et de la conformité des plateformes avec le droit de l’UE et le droit national.

Le gouvernement britannique prévoit des mesures qui pourraient exposer les dirigeants du secteur technologique à des poursuites pénales, y compris des peines d’emprisonnement, si leurs entreprises ne suppriment pas les images intimes non consenties lorsque les régulateurs l’exigent. Cette initiative s’appuie sur les obligations existantes qui imposent déjà aux plateformes de retirer ce type de contenu dans des délais stricts sous peine de sanctions importantes, notamment des amendes pouvant atteindre 10 % du chiffre d’affaires mondial, voire le blocage des services. Cette dernière mesure va plus loin : au lieu de s’appuyer uniquement sur des sanctions à l’encontre des entreprises, elle introduit une responsabilité pénale personnelle au niveau des dirigeants. Ce type de responsabilité est susceptible d’accélérer la mise en conformité d’une manière que les sanctions financières seules n’ont pas permis, et pourrait servir d’exemple à d’autres juridictions. Cette politique s’inscrit dans le cadre d’un renforcement plus large du dispositif britannique de sécurité en ligne, motivé par des préoccupations persistantes concernant la pornographie de vengeance et la prolifération rapide d’images intimes générées par l’IA.

Ce mois-ci dans la gouvernance de l’IA

Plans et initiatives nationaux

Inde. L’Inde a mis en place un comité d’experts en technologie et en politique sous l’égide du ministère de l’Électronique et des Technologies de l’information afin d’influencer l’élaboration du cadre de gouvernance de l’IA du pays et de conseiller le nouveau groupe sur la gouvernance et l’économie de l’IA. Réunissant des experts du gouvernement, du monde universitaire, de l’industrie et des politiques publiques, cet organisme a pour mission de traduire les questions techniques et réglementaires en constante évolution en orientations pratiques, apportant ainsi une approche plus structurée et adaptative de la gouvernance de l’IA, alignée sur les priorités économiques et sociales de l’Inde.

Afrique du Sud. L’Afrique du Sud a retiré son projet de politique nationale sur l’IA après avoir découvert que le document contenait de fausses citations générées par l’IA, ce qui compromettait la crédibilité du cadre proposé. Le gouvernement a déclaré que cette erreur était due à un manque de vérification des références et a souligné qu’une surveillance humaine plus stricte était nécessaire dans les processus politiques impliquant des outils d’IA. Ce retrait retarde les projets de création de nouvelles institutions de gouvernance de l’IA et de mesures incitatives, et la politique va désormais être réécrite.

Souveraineté

Royaume-Uni. La secrétaire d’État britannique à la Science, à l’Innovation et à la Technologie, Liz Kendall, a déclaré que le gouvernement prévoyait de soutenir les atouts britanniques dans les segments de la chaîne technologique de l’IA où le Royaume-Uni peut développer un véritable levier stratégique. Mme Kendall a rejeté l’idée d’un isolationnisme technologique et a au contraire défendu une souveraineté britannique en matière d’IA : réduire les dépendances excessives, soutenir les entreprises nationales grâce à un fonds souverain pour l’IA de 500 millions de livres sterling, et lancer en juin 2026 un nouveau plan consacré au matériel IA afin de gagner des parts de marché dans le secteur des puces électroniques. Elle a également plaidé pour une coopération avec d’autres puissances intermédiaires, notamment pour définir les normes encadrant le déploiement de l’IA.

Le gouvernement a également lancé un fonds souverain dédié à l’IA, de 500 millions de livres sterling, afin d’accélérer le développement des start-ups nationales spécialisées en IA et de renforcer l’autonomie technologique du pays. Cette initiative combine des investissements directs en capital avec un accès aux infrastructures informatiques nationales, des procédures accélérées d’obtention de visas pour les talents internationaux ainsi que des mécanismes facilitant l’accès des entreprises aux marchés publics et aux services publics. Le programme cible les entreprises en phase de démarrage ou de croissance dans des domaines tels que l’infrastructure d’IA, les sciences de la vie et l’informatique de pointe, avec pour objectif explicite de garantir que les entreprises à fort potentiel se développent et restent ancrées au Royaume-Uni plutôt que de délocaliser à l’étranger.

Papouasie-Nouvelle-Guinée. Le gouvernement a publié de nouvelles lignes directrices sur l’IA et la souveraineté des données, définissant des principes visant à garantir que les ressources nationales en matière de données restent sous contrôle national. Ce cadre met l’accent sur la gouvernance du stockage, du traitement et des transferts transfrontaliers des données, en particulier lorsqu’il s’agit de données du secteur public ou de données sensibles.

Russie. La Russie propose un projet de cadre réglementaire sur l’IA qui officialiserait la supervision du développement et du déploiement de l’IA, s’inscrivant ainsi dans le cadre d’efforts plus larges visant à renforcer la souveraineté numérique et le contrôle de l’État sur les technologies émergentes. Les propositions se concentrent sur la gestion des risques, les normes nationales et la réduction de la dépendance vis-à-vis des systèmes d’IA étrangers, tout en soutenant l’innovation nationale. Cette initiative s’inscrit dans la stratégie plus large de Moscou visant à renforcer le contrôle sur les infrastructures numériques et les flux de données transfrontaliers.

Partenariats

Corée du Sud – France. La Corée du Sud et la France renforcent leur coopération par le biais d’un nouveau partenariat stratégique en matière d’IA et de technologie, visant à consolider la recherche conjointe, la collaboration industrielle et la normalisation dans le domaine des technologies émergentes. Cette initiative s’inscrit dans un effort plus large visant à harmoniser les capacités en matière de semi-conducteurs, d’infrastructures de données et de calcul avancé, tout en positionnant les deux pays de manière plus compétitive sur la scène mondiale de l’IA.

L’UE et le Maroc. La Commission européenne et le Maroc ont lancé un dialogue numérique afin d’approfondir leur coopération stratégique sur les technologies émergentes, la transformation numérique et le développement axé sur l’innovation. Ciblées sur l’IA, les infrastructures numériques, le soutien aux start-ups, la collaboration en matière de recherche et le renforcement des écosystèmes d’IA, cette initiative vise à faire des technologies numériques des moteurs du progrès économique et social. Une plus grande interopérabilité des services publics numériques et un échange de connaissances élargi sont également au cœur de ce partenariat, reflétant un intérêt commun pour une gouvernance numérique plus connectée, plus efficace et plus inclusive.

Juridique

États-Unis. Une cour d’appel fédérale de Washington, D.C. a refusé de bloquer l’inscription d’Anthropic sur la liste noire du Pentagone pour des raisons de sécurité nationale, permettant ainsi à cette mesure de rester en vigueur pendant que le litige se poursuit. Cette décision contraste avec celle d’un juge californien qui avait auparavant bloqué une partie de la mesure gouvernementale, soulignant une divergence judiciaire croissante face à cette initiative sans précédent.

Paraguay. Le Paraguay a adopté de nouvelles règles relatives à l’utilisation de l’IA dans ses tribunaux, avec le soutien de l’UNESCO, marquant ainsi une avancée notable dans la gouvernance judiciaire de l’IA. Ce cadre, approuvé par la Cour suprême de justice, limite l’IA à un rôle de soutien dans le traitement des données, la gestion de l’information et l’aide à la prise de décision, tout en exigeant une supervision humaine, la transparence, la responsabilité et la divulgation lorsque les outils d’IA influencent les processus judiciaires. Ces règles alignent l’approche du Paraguay sur les recommandations de l’UNESCO concernant l’IA dans les tribunaux et soulignent une tendance plus large vers un déploiement de l’IA fondé sur les droits et axé sur la confiance au sein des institutions publiques.

Belgique. L’autorité belge chargée de la protection des données a publié une nouvelle brochure d’information intitulée « L’impact de l’intelligence artificielle (IA) sur la vie privée », qui fournit des conseils sur les risques tels que les biais, les violations de la vie privée et l’utilisation abusive des systèmes d’IA générative. Ce document vise à sensibiliser les organisations et le public, et à favoriser le respect des cadres européens en matière de protection des données et de gouvernance de l’IA.

Sécurité et sûreté

L’UE. Les États membres de l’UE et les députés du Parlement européen n’ont pas réussi à trouver un accord sur la révisions de la loi européenne sur l’intelligence artificielle, après 12 heures de négociations sur les modifications proposées dans le cadre du paquet « Omnibus numérique » de la Commission. Les désaccords portaient principalement sur la question de savoir si les secteurs déjà couverts par les réglementations existantes en matière de produits et de sécurité devaient être exemptés de certaines parties du cadre réglementaire sur l’IA. Les législateurs ont averti que cette nouvelle impasse risquait de créer une insécurité juridique pour les entreprises qui se préparent déjà à se conformer à la réglementation, tandis que les groupes de défense de la vie privée et de la société civile ont mis en garde contre le fait que les assouplissements proposés pourraient affaiblir les garanties fondamentales. Les discussions reprendront toutefois en mai.

Kazakhstan. Le Kazakhstan a mis en place des audits obligatoires pour les systèmes d’IA à haut risque, exigeant des développeurs qu’ils obtiennent une évaluation d’audit positive avant que leurs systèmes puissent être classés comme « fiables » par les autorités sectorielles. Le gouvernement publiera et mettra régulièrement à jour des listes officielles de systèmes approuvés, sur la base de demandes comprenant des documents relatifs à la propriété, aux fonctionnalités et aux conditions d’utilisation, examinées dans des délais stricts. Cette mesure vise à instaurer la confiance et à normaliser les meilleures pratiques en matière de déploiement de l’IA, marquant ainsi une approche plus structurée et axée sur la conformité de la gouvernance de l’IA à haut risque.

Nouvelle-Zélande, Royaume-Uni, Singapour. Le Centre national de cybersécurité de Nouvelle-Zélande, le Centre national de cybersécurité du Royaume-Uni et l’Agence de cybersécurité de Singapour ont publié des avertissements coordonnés indiquant que l’IA de pointe est en train de redessiner le paysage des cybermenaces en abaissant les barrières aux attaques sophistiquées, en accélérant la découverte des vulnérabilités et en réduisant le délai entre la divulgation et l’exploitation. Tous trois soulignent la nature à double usage de l’IA, exhortant les organisations à réévaluer les modèles de risque obsolètes et à donner la priorité à l’application rapide de correctifs, à la surveillance continue, à des contrôles d’identité et d’accès renforcés, ainsi qu’à la réduction des surfaces d’attaque afin de contrer les cybermenaces de plus en plus automatisées et rapides dans les secteurs public et privé.

États-Unis. Les responsables américains de la cybersécurité envisagent de ramener à trois jours seulement le délai de correction des failles activement exploitées, invoquant la rapidité croissante à laquelle les systèmes d’IA peuvent identifier et exploiter les vulnérabilités. Ce changement proposé s’appliquerait dans un premier temps aux agences civiles fédérales, mais pourrait redéfinir les attentes de base en matière de réponse aux incidents au sein du gouvernement et des infrastructures critiques. Les agences font valoir que les cycles de correctifs traditionnels ne sont plus compatibles avec les délais d’exploitation actuels, tandis que le secteur met en garde contre le fait que des délais aussi raccourcis pourraient dépasser la capacité des environnements informatiques complexes et hérités.

Dans le même temps, Washington fait discrètement marche arrière dans son bras de fer avec Anthropic. La Maison Blanche rédige actuellement des directives exécutives qui permettraient aux agences fédérales de travailler à nouveau avec Anthropic, bien que l’entreprise ait précédemment été qualifiée de risque pour la chaîne d’approvisionnement par le Pentagone. Ce revirement reflète des divisions internes : alors que certains secteurs de la défense restent méfiants, d’autres considèrent que l’exclusion de modèles de pointe comme Mythos serait stratégiquement coûteuse.

Mythos. Anthropic a ouvert une enquête après qu’un petit groupe d’utilisateurs a obtenu un accès non autorisé à son puissant modèle d’IA Mythos via l’environnement d’un sous-traitant tiers. Cet accès aurait eu lieu au moment même où l’entreprise commençait à déployer une version préliminaire limitée du modèle auprès d’organisations sélectionnées dans le cadre du projet Glasswing. Les utilisateurs non autorisés auraient opéré via un groupe Discord privé, en utilisant diverses tactiques, notamment l’accès par le biais de sous-traitants et des outils de renseignement open source, pour accéder au système. Mythos avait été intentionnellement restreint en raison de sa capacité à accélérer les cyberattaques et n’avait été fourni qu’à un nombre limité de partenaires, mais il semble avoir fait l’objet d’une fuite presque immédiate via l’écosystème des partenaires plutôt que par une violation directe. La période pendant laquelle les capacités de Mythos resteront contenues pourrait s’avérer bien plus courte que prévu.

Gouvernance des contenus

Chine. L’Administration chinoise du cyberespace a mis en garde plusieurs plateformes appartenant à ByteDance, notamment CapCut, Catbox et le système d’IA Dreamina, pour ne pas avoir correctement signalé les contenus générés par l’IA et les contenus synthétiques. L’Administration chinoise du cyberespace a déclaré que des inspections avaient révélé des violations des réglementations en matière de cybersécurité et d’IA générative, ce qui a donné lieu à des mesures coercitives telles que des rectifications obligatoires, des avertissements et des sanctions disciplinaires à l’encontre du personnel responsable.

Développement

Ghana. Le ministère ghanéen de la Communication, des Technologies numériques et des Innovations a lancé un programme de renforcement des capacités en matière d’IA dans le secteur public, en collaboration avec le gouvernement japonais et le Programme des Nations Unies pour le développement. Ce programme vise à doter les fonctionnaires de connaissances sur l’IA et ses applications dans la gouvernance. Il met l’accent sur l’amélioration de la prise de décision et de la prestation de services, en s’appuyant sur l’expérience des Nations Unies et du Japon.

UNESCO – Amérique latine et Caraïbes. L’UNESCO a lancé un observatoire régional de l’IA dans l’éducation pour l’Amérique latine et les Caraïbes, conçu pour soutenir l’élaboration de politiques fondées sur des données factuelles et suivre l’impact de l’IA sur les systèmes éducatifs. Cette initiative vise à renforcer les capacités, à partager les meilleures pratiques et à orienter l’intégration responsable des outils d’IA dans les écoles et les environnements d’apprentissage.

UNESCO–Oxford. L’UNESCO et l’université d’Oxford ont lancé une formation mondiale sur l’IA destiné aux tribunaux. Ce programme forme les juges et les professionnels du droit à évaluer les outils algorithmiques, à identifier les biais et à garantir le respect des normes en matière de droits de l’Homme dans des procédures judiciaires de plus en plus numérisées. Il présente des cadres pratiques pour évaluer les résultats de l’IA dans des contextes juridiques, en mettant fortement l’accent sur le maintien de l’indépendance judiciaire, de la transparence et de la responsabilité à mesure que l’IA s’intègre dans les systèmes de traitement des preuves et d’aide à la décision.

Commonwealth. Le Secrétariat du Commonwealth a lancé un programme de renforcement des capacités sur l’utilisation de l’IA dans la gestion électorale, formant les responsables électoraux des États membres à la manière dont les outils d’IA peuvent contribuer à l’éducation des électeurs, à l’efficacité administrative et à l’analyse des données tout en préservant l’intégrité électorale. Cette initiative se concentre sur les applications pratiques de l’IA dans les processus électoraux, y compris les risques tels que la désinformation, les biais et l’automatisation de fonctions sensibles d’aide à la décision. Elle souligne que l’IA doit rester un outil d’assistance plutôt qu’un substitut dans les processus démocratiques, le contrôle humain étant considéré comme essentiel pour maintenir la confiance, la légitimité et la responsabilité dans les élections.

Australie. Dans le cadre de sa stratégie nationale pour la main-d’œuvre dans le domaine de l’IA, l’Australie développe des programmes de perfectionnement ciblés destinés aux apprenants et aux travailleurs afin de combler les déficits de compétences structurels créés par les mutations du marché du travail induites par l’IA. Cette approche donne la priorité à l’intégration de la culture de l’IA dans les parcours éducatifs et professionnels, parallèlement à des formations en lien avec les employeurs pour soutenir l’adaptation dans les secteurs fortement exposés. Elle considère l’IA comme une technologie à usage général nécessitant une reconversion continue plutôt qu’une formation ponctuelle, l’attention politique se portant sur l’inclusion, l’accompagnement de la transition et l’alignement entre les systèmes éducatifs et les exigences de l’économie numérique émergente.

Pakistan. Le Pakistan a approuvé la création d’une Autorité chargée de l’éducation à l’IA, parallèlement à des projets d’écoles virtuelles. Ces réformes visent à déployer à grande échelle des systèmes d’apprentissage basés sur l’IA, à soutenir la mise en œuvre d’un enseignement personnalisé et à harmoniser les programmes numériques entre les régions. Cette initiative s’inscrit dans le cadre d’efforts plus larges visant à moderniser le secteur de l’éducation, à renforcer l’accès au numérique et à développer les capacités nationales pour l’adoption de l’IA dans l’enseignement public, tout en remédiant aux disparités dans les résultats d’apprentissage grâce à des modèles d’enseignement assistés par la technologie.

Quand l’IA écrit les règles : comment éviter que des lois fictives régissent la réalité

Le mois dernier, l’Afrique du Sud a dévoilé son premier projet de politique nationale en matière d’IA, visant à positionner le pays comme un chef de file du continent en matière d’innovation. Ce plan prévoyait la création de nouvelles institutions ambitieuses : une Commission nationale de l’IA, un Conseil d’éthique, ainsi que des allègements fiscaux pour encourager la collaboration avec le secteur privé.

Mais quelques jours plus tard, la liesse s’est transformée en embarras.

Selon Reuters, le gouvernement sud-africain a été contraint de retirer le projet après que des examinateurs ont découvert une faille fatale : le projet était truffé de fausses sources et de citations imaginaires. Les recherches étayant la stratégie du pays en matière d’IA avaient probablement été générées par une IA.

Il ne s’agit pas d’une simple faute de frappe. L’IA a inventé de toutes pièces des politiques et des ressources pour les étayer. Cela n’a rien de surprenant, car les grands modèles linguistiques sont des machines de prédiction sophistiquées, et non des fournisseurs de faits vérifiés. Même lorsqu’ils sont faux, les textes peuvent paraître parfaitement corrects et légitimes.

Le ministre sud-africain des Communications et des Technologies numériques, Solly Malatsi, a reconnu cet échec avec une honnêteté rafraîchissante :

« L’explication la plus plausible est que des citations générées par l’IA ont été incluses sans vérification appropriée. Cela n’aurait pas dû se produire. »

Il a noté que cette erreur « a compromis l’intégrité et la crédibilité du projet de politique ».

Pourquoi est-ce important ? Nous ne mettons pas l’Afrique du Sud sous les feux de la rampe pour l’embarrasser. Nous mettons en lumière le problème des lois générées par l’IA. L’incident survenu en Afrique du Sud n’est pas une exception. Alors que les décideurs politiques s’empressent de suivre le rythme de la technologie, nous voyons de plus en plus d’exemples de réglementations rédigées par l’IA soumises pour examen. Par exemple, aux États-Unis, un juge fédéral de Californie a sanctionné deux cabinets d’avocats pour avoir soumis un mémoire juridique contenant une fausse citation générée par l’IA.

Le problème n’est pas que l’IA soit utilisée. Le danger réside dans la manière dont elle est utilisée.

Les documents juridiques et les politiques exigent précision, fondement et contextualisation. Les modèles d’IA génériques échouent souvent sur ces trois points :

Manque de précision : l’IA fournit fréquemment des réponses vagues et génériques à des questions juridiques spécifiques. Les lois ont besoin de définitions précises et solides ; l’IA préfère les conjectures probabilistes.
Absence de fondement : la plupart des modèles d’IA ne peuvent pas fournir de lien vérifiable vers la phrase exacte d’une loi ou d’un règlement. Souvent, ils confondent les juridictions entre les pays et les juridictions.
Absence de contexte : l’IA manque souvent du contexte politique, social ou historique spécifique d’une politique ou d’une réglementation. Le contexte temporel fait également défaut, ce qui montre comment les questions juridiques ont évolué au cours de la rédaction et des négociations.

Comment résoudre ce problème (sans interdire l’IA). La solution réside dans une double approche : développer une IA institutionnelle et améliorer la culture numérique en matière d’IA.

Si l’Afrique du Sud avait disposé d’une IA institutionnelle ancrée dans les connaissances et les réalités locales, une telle dérive aurait pu être évitée. De plus, l’IA serait alors un outil fiable et utile reflétant le contexte thématique et temporel de l’élaboration des politiques et de la rédaction des lois.

Mais surtout, nous devons développer les compétences en IA chez les décideurs politiques. Cela nécessite un changement de pédagogie. Nous ne pouvons pas nous contenter d’apprendre aux décideurs politiques à utiliser l’IA ; ils doivent comprendre comment elle fonctionne.

Comme l’a déclaré le ministre Malatsi :

« Cette erreur inacceptable prouve pourquoi une surveillance humaine vigilante de l’utilisation de l’intelligence artificielle est essentielle. C’est une leçon que nous tirons avec humilité. »

Si nous ne parvenons pas à mettre au point des outils d’IA précis et fiables et à former les décideurs politiques à les utiliser correctement, nous n’aurons pas seulement de fausses citations dans un projet de loi. Nous aurons de fausses lois régissant la vie de personnes réelles.

Lisez l’article original « Quand l’IA rédige les règles : comment éviter que de fausses lois régissent la réalité » du Dr Jovan Kurbalija.as : le combat pour l’avenir du savoir humain.».

Concevoir notre avenir

Le mois d’avril a vu déferler une nouvelle vague de prédictions très médiatisées. Si certains pourraient être tentés de les rejeter comme de simples effets d’annonce, il existe une bonne raison de ne pas le faire. Ces idées émanent de personnes qui non seulement développent les plateformes ou les technologies sur lesquelles nous dépendons tant, mais qui investissent également des capitaux pour transformer en réalité leur vision de ce à quoi devrait ressembler l’avenir. Lorsque les « leaders technologiques » lancent leurs idées, ils commencent à orienter les ressources du monde réel et les débats réglementaires. Et c’est là que réside le danger silencieux : concevoir l’avenir sans consultation significative du public.

Le cofondateur de Palantir, Alex Karp, et le directeur des affaires générales de Palantir, Nicholas W. Zamiska, ont publié une série de 22 propositions tirées de leur prochain ouvrage, Technological Republic. Ce livre n’est pas passé inaperçu. Les détracteurs l’ont qualifié de « technofascisme » et de « ce que le mal tweeterait ».

Leur vision s’articule autour du devoir, de la force brute et du scepticisme envers la culture démocratique moderne. Ils soutiennent que la Silicon Valley a une dette morale envers le pays qui a rendu son essor possible, et que l’élite des ingénieurs a l’obligation positive de participer à la défense nationale. Ils remettent en question l’armée de volontaires, suggérant que le service national devrait être un devoir universel afin que la prochaine guerre implique un risque partagé. Le soft power et la rhétorique grandiloquente, écrivent-ils, se sont révélés insuffisants. Les sociétés libres ont besoin de force militaire, et en ce siècle, celle-ci reposera sur les logiciels.

En ce qui concerne les armes basées sur l’IA, Karp et Zamiska sont catégoriques : elles seront développées, indépendamment des débats occidentaux. La seule question est de savoir par qui et dans quel but. Les auteurs prennent également la défense d’Elon Musk contre ce qu’il considère être des moqueries culturelles, en faisant valoir que nous devrions applaudir ceux qui tentent d’agir là où le marché a failli à sa mission. Dans le même temps, ils rejettent ce qu’ils appellent le « pluralisme vide », insistant sur le fait que toutes les cultures ne sont pas aussi productives les unes que les autres et que l’intolérance de l’élite envers les croyances religieuses est un signe de fermeture intellectuelle.

Ce que Kapr et Zamiska ne proposent pas, ce sont des mesures de politique économique. Leur république technologique s’articule autour de la sécurité et de la puissance technologique, et non de la redistribution. L’État existe pour être défendu. L’individu existe pour servir.

À peu près à la même époque, OpenAI a publié son propre document d’orientation, intitulé « Industrial Policy for the Intelligence Age ». Ce document est plus long, d’un ton plus modéré, et regorge d’expressions telles que « fonds souverain » et « droit à l’IA ». Il appelle à un débat démocratique sur la politique industrielle, la réglementation, l’éthique et l’économie en matière d’IA.

Le document d’OpenAI part d’un constat différent. La superintelligence — ces systèmes d’IA capables de surpasser les humains les plus brillants, même lorsque ces derniers sont assistés par l’IA — est en train d’arriver. OpenAI estime que les forces du marché ne peuvent à elles seules gérer cette transition. En établissant des parallèles avec l’ère progressiste et le New Deal, l’entreprise propose une collaboration public-privé ambitieuse.

Sur le plan économique, cela inclut notamment de donner aux travailleurs une voix officielle sur la manière dont l’IA est déployée sur les lieux de travail, des micro-subventions pour aider les travailleurs à devenir des entrepreneurs axés sur l’IA, un droit à l’IA en tant qu’accès fondamental comparable à l’alphabétisation ou à l’électricité, le transfert de la fiscalité des salaires vers les plus-values et le travail automatisé, la création d’un fonds de richesse publique pour donner aux citoyens une participation directe à la croissance tirée par l’IA, et la conversion des gains d’efficacité en semaines de travail plus courtes ou en meilleures prestations sociales.

Sur le plan de la résilience, OpenAI propose des systèmes de sécurité contre les risques cybernétiques et biologiques, une pile de confiance en IA pour la vérification, des régimes d’audit pour les modèles de pointe, des guides de confinement des modèles pour l’IA dangereuse, ainsi que des garde-fous pour l’utilisation par les gouvernements. L’entreprise reconnaît ne pas détenir toutes les réponses et invite à lui faire part de vos commentaires.

Similitudes et différences. Là où Karp et Zamiska parlent de devoir et de guerre, OpenAI parle de transitions et de dispositifs de sécurité. Pourtant, tous deux rejettent l’ordre politique actuel, qu’ils jugent inadéquat. Tous deux considèrent la technologie comme le principal vecteur de pouvoir. Et tous deux proposent de nouvelles formes d’obligation : le service national dans un cas, un droit à l’IA et des avantages transférables dans l’autre.

Pris ensemble, ces deux documents ne sont pas des manifestes opposés. Ce sont des dialectes différents d’un même langage émergent : les leaders technologiques ne se considèrent plus comme des fabricants d’outils. Ils se considèrent comme des concepteurs d’institutions. Et une bataille judiciaire entre Elon Musk et Sam Altman est sur le point de déterminer dans quelle mesure leurs promesses initiales sont réellement exécutoires.

Des promesses, encore des promesses. Elon Musk poursuit Sam Altman en justice pour déterminer si OpenAI s’est détournée de manière frauduleuse de sa mission initiale à but non lucratif. M. Musk soutient qu’il a été induit en erreur et que la direction d’OpenAI a renoncé à sa promesse de servir l’humanité, s’orientant plutôt vers la commercialisation par le biais de partenariats et de produits tels que ChatGPT. Il cherche à écarter Altman et le président Greg Brockman, à imposer des changements structurels à la gouvernance d’OpenAI et à obtenir potentiellement jusqu’à 150 milliards de dollars de dommages-intérêts pour sa branche à but non lucratif. OpenAI rejette cette version des faits, qualifiant l’affaire de litige concurrentiel : selon eux, Musk n’a soulevé des objections qu’après le succès d’OpenAI et l’émergence de sa propre entreprise d’IA, xAI, qui a déposé une demande d’introduction en bourse. Selon certaines rumeurs, OpenAI envisagerait elle-même une introduction en bourse fin 2026 ou en 2027. Le tribunal devra examiner les premiers e-mails, les discussions sur le financement et les interprétations contradictoires de ce que les termes « ouvert » et « à but non lucratif » étaient censés signifier.

Si le tribunal estime que l’orientation vers le profit a violé les principes fondateurs, de nombreuses organisations hybrides similaires pourraient devoir se restructurer. Si le modèle actuel est maintenu, cela consolidera la réalité selon laquelle la logique du marché et les intérêts commerciaux sont les moteurs du développement de l’IA. La mise au point et l’exploitation d’une IA avancée étant coûteuses, les entreprises ont besoin de barèmes tarifaires pour couvrir leurs coûts et réaliser des bénéfices. Et comme les modèles et l’infrastructure sous-jacents constituent des atouts concurrentiels précieux, les entreprises sont incitées à fidéliser les utilisateurs et à limiter la divulgation d’informations afin de conserver leur avantage. Cela signifie que les utilisateurs pourraient être confrontés à un accès plus différencié, à un verrouillage plus fort sur les plateformes et à une visibilité réduite sur le fonctionnement des systèmes.

Que pourraient donc faire les sociétés ? Karp, Zamiska et OpenAI partagent un postulat rarement énoncé ouvertement : l’ordre juridique et politique existant est trop lent ou trop confus pour gérer les technologies qui émergent actuellement.

Si l’on part du principe qu’ils ont ne serait-ce que partiellement raison, la solution ne peut consister à confier le pouvoir de conception aux mêmes entreprises qui tirent profit de ces technologies. Trois mesures méritent d’être envisagées.

Premièrement, séparer la conception des politiques de la stratégie d’entreprise. Toute entreprise détenant d’importants contrats publics dans des domaines tels que la défense, la santé ou le contrôle des frontières ne devrait pas être à l’origine des politiques utilisées pour réglementer ses propres activités.

Deuxièmement, codifier la responsabilité. Si les développeurs d’IA revendiquent des missions d’intérêt public, ces revendications doivent reposer sur des fondements juridiques et réglementaires, et non pas simplement sur une image de marque. L’affaire Musk-OpenAI pourrait accélérer ce processus, mais les décideurs politiques ne peuvent pas externaliser cette tâche aux tribunaux.

Troisièmement, élargir la participation. L’appel d’OpenAI à la contribution du public va dans la bonne direction, mais les mécanismes comptent. Sans une inclusion significative — impliquant les travailleurs, la société civile et les petites économies —, la participation risque de devenir procédurale plutôt que substantielle.

Nous ne sommes pas sur le point de nous réveiller d’un jour à l’autre dans une république technologique. Mais il est déjà clair que les oligarques de la technologie ne se contentent plus de créer des produits ; ils façonnent des ordres politiques et sociaux. Les sociétés modernes devront déterminer quel type d’ordre juridique et politique est nécessaire et comment faire face au pouvoir croissant des entreprises technologiques et de leurs dirigeants.

Claude Mythos Preview établit une nouvelle référence en matière de capacités de l’IA et soulève des questions de gouvernance

Le 7 avril 2026, Anthropic a annoncé le lancement de Claude Mythos Preview, son modèle d’IA le plus performant à ce jour, tout en prenant la décision explicite de ne pas le rendre accessible au public. Claude Mythos Preview est un modèle de pointe polyvalent, non commercialisé, qui, selon les propres termes d’Anthropic, met en évidence une réalité frappante : les modèles d’IA ont atteint un niveau de compétence en programmation tel qu’ils sont capables de surpasser tous les humains, à l’exception des plus chevronnés, dans la détection et l’exploitation des failles logicielles.

Les benchmarks publiés par Anthropic montrent que Mythos Preview a obtenu un score de 93,9 % au test SWE-bench Verified, de 97,6 % à l’évaluation mathématique de l’USAMO 2026, et a largement surpassé tous les modèles précédemment publiés lors d’évaluations spécifiques à la cybersécurité. Le score SWE-bench Verified est environ le double de l’état de l’art de 2024 et a été obtenu dans un contexte agentique, où le modèle a résolu de manière autonome de véritables problèmes d’ingénierie logicielle à partir de bases de code de production.

Lors de l’évaluation USAMO 2026, Mythos Preview a obtenu un score supérieur de 55 points de pourcentage à celui d’Opus 4.6, qui a obtenu 42,3 %. Sur GPQA Diamond, un indice de référence de raisonnement scientifique de niveau universitaire, Mythos Preview a obtenu un score de 94,6 %. Sur Terminal-Bench 2.0, qui évalue les compétences en administration système et en ligne de commande, il a obtenu un score de 82,0 %, soit une avance de 16,6 points sur Opus 4.6. Sur le benchmark de cybersécurité Cybench, le modèle a obtenu un score de 100 % dès le premier essai, ce qui rend ce benchmark inutilisable comme évaluation discriminante.

Capacités en matière de cybersécurité

La décision de ne pas rendre Mythos Preview accessible au public est liée à des préoccupations concernant ses capacités avancées, en particulier dans des domaines à haut risque tels que la cybersécurité, ainsi qu’à des considérations plus générales liées à la sécurité et aux risques d’utilisation abusive.

Il convient de noter que ces capacités ne sont pas le résultat d’un apprentissage ciblé. Anthropic n’a pas explicitement formé Mythos Preview pour qu’il dispose de ces capacités. Elles sont apparues comme une conséquence indirecte des améliorations générales apportées au code, au raisonnement et à l’autonomie. Les mêmes améliorations qui rendent le modèle nettement plus efficace pour corriger les vulnérabilités le rendent également nettement plus efficace pour les exploiter.

Au cours de tests internes, Mythos Preview a identifié des milliers de failles « zero-day » sur tous les principaux systèmes d’exploitation et navigateurs web, ainsi que sur d’autres logiciels critiques, dont beaucoup étaient de gravité élevée et n’avaient pas été détectées depuis des années. Des ingénieurs d’Anthropic n’ayant reçu aucune formation officielle en sécurité pouvaient demander à Mythos de trouver des failles permettant l’exécution de code à distance pendant la nuit et disposer d’un outil d’exploitation complet et fonctionnel le lendemain matin. Cette dimension d’accessibilité soulève un problème de gouvernance distinct. Traditionnellement, les cyberattaques sophistiquées nécessitaient des équipes hautement qualifiées, une planification approfondie et une expertise technique pointue. Les modèles dotés de ces capacités peuvent réduire considérablement ces barrières, y compris pour les petits acteurs étatiques et non étatiques.

Anthropic n’a divulgué qu’une fraction de ce qu’elle affirme avoir découvert lors de ses tests internes. Plus de 99 % des failles découvertes par Mythos n’avaient toujours pas fait l’objet d’un correctif au moment de l’annonce du 7 avril.

Projet Glasswing

Anthropic a lancé le projet Glasswing en tant que mécanisme d’accès structuré permettant d’utiliser Claude Mythos Preview à des fins de cybersécurité défensive. Cette initiative rassemble Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, la Linux Foundation, Microsoft, NVIDIA et Palo Alto Networks en tant que partenaires de lancement, l’accès étant également étendu à plus de 40 autres organisations qui développent ou gèrent des infrastructures logicielles critiques.

Les partenaires du projet Glasswing auront accès à Claude Mythos Preview pour détecter et corriger les failles de leurs systèmes fondamentaux. Les travaux devraient se concentrer sur la détection locale des vulnérabilités, les tests de type « boîte noire » des binaires, la sécurisation des terminaux et les tests d’intrusion. Anthropic s’engage à fournir jusqu’à 100 millions de dollars de crédits d’utilisation pour Mythos Preview dans le cadre de ces efforts. À l’issue de la période initiale de prévisualisation de la recherche, l’accès au modèle sera proposé aux participants au tarif de 25 dollars par million de jetons d’entrée et de 125 dollars par million de jetons de sortie sur l’API Claude, Amazon Bedrock, Vertex AI de Google Cloud et Microsoft Foundry.

Anthropic a également fait don de 2,5 millions de dollars à Alpha-Omega et à l’OpenSSF par l’intermédiaire de la Linux Foundation, et de 1,5 million de dollars à l’Apache Software Foundation afin de permettre aux responsables de la maintenance des logiciels open source de s’adapter à l’évolution du paysage de la cybersécurité.

Anthropic s’est engagée à rendre publiques ses conclusions, dans un délai de 90 jours, de ce qu’elle a appris, ainsi que des failles corrigées et des améliorations apportées pouvant être divulguées. L’entreprise a également l’intention de collaborer avec des organisations de sécurité de premier plan afin de formuler des recommandations pratiques couvrant, entre autres, les processus de divulgation des vulnérabilités, les processus de mise à jour des logiciels, la sécurité de l’open source et de la chaîne d’approvisionnement, ainsi que l’automatisation des correctifs.

Anthropic a déclaré que le projet Glasswing constituait un point de départ et qu’à moyen terme, un organisme tiers indépendant réunissant des organisations des secteurs privé et public pourrait être le cadre idéal pour poursuivre les travaux sur des projets de cybersécurité à grande échelle.

Le projet Glasswing soulève une question de gouvernance pour le secteur, car les systèmes d’IA dotés de capacités cybernétiques peuvent devenir à la fois des outils de sécurité utiles et une source de risque d’utilisation abusive. La structure du projet Glasswing révèle également des tensions, car elle concentre plusieurs rôles, notamment la découverte, la coordination de la divulgation et le contrôle des capacités, au sein d’une seule organisation. Des entités telles qu’Anthropic et les principaux fournisseurs de cloud contrôlent des composants critiques de l’écosystème Glasswing, ce qui soulève des questions de pouvoir et de gouvernance qui, pour les institutions financières en particulier, se traduisent par un risque systémique.

Nous avons également abordé le projet Glasswing et ses implications dans notre newsletter hebdomadaire début avril.

Dimensions géopolitiques

Claude Mythos a attiré l’attention sur les dimensions concurrentielles et géopolitiques du développement de l’IA de pointe. Les partenaires de lancement du projet Glasswing excluent OpenAI, le rival d’Anthropic, qui aurait environ six mois de retard sur Anthropic dans le développement d’un modèle doté de capacités cyberoffensives comparables.

Des responsables politiques de haut niveau ont replacé Mythos dans le contexte plus large de la concurrence entre les entreprises occidentales d’IA et l’écosystème chinois de l’IA en pleine évolution, avec des implications pour la sécurité nationale, l’adoption par les entreprises et le leadership technologique. Un chercheur en sécurité a qualifié une fuite simultanée de code source chez Anthropic d’accélérateur géopolitique, soulignant que de telles divulgations réduisent le délai dont disposent les adversaires pour reproduire les avantages technologiques dont bénéficient actuellement les laboratoires occidentaux.

De nombreuses organisations de défense continuent de dépendre de logiciels et d’infrastructures hérités qui n’ont pas été conçus pour faire face à des menaces alimentées par l’IA. Des modèles capables d’identifier de manière autonome des failles cachées dans d’anciens codes informatiques pourraient révéler des vulnérabilités au sein de réseaux de défense critiques à travers le monde. La difficulté à contenir ces capacités à l’échelle géopolitique se reflète dans les modes d’utilisation observés. Le fait de restreindre l’accès au niveau des laboratoires ne garantit pas réellement un confinement entre juridictions lorsque les mêmes modèles sous-jacents restent accessibles via des infrastructures cloud couvrant plusieurs pays et environnements réglementaires.

Les limites de la gouvernance volontaire de l’IA

L’affaire Claude Mythos a clarifié, avec une grande précision, ce que la gouvernance volontaire de l’IA peut et ne peut pas accomplir. Un laboratoire responsable peut prendre la décision unilatérale de ne pas commercialiser un système dangereux. Il peut soutenir la divulgation coordonnée des vulnérabilités, impliquer les gouvernements de manière proactive et produire une documentation publique détaillée sur les capacités et les risques d’un modèle. Tout cela s’est produit avec Mythos et représente un progrès significatif par rapport à l’environnement de gouvernance d’il y a quelques années.

Ce que les cadres volontaires ne peuvent pas faire, c’est contraindre des concurrents qui opèrent selon des hypothèses différentes. La version 3.0 du RSP d’Anthropic le reconnaît directement en supprimant l’engagement de ne pas commercialiser de modèles dangereux si un autre laboratoire commercialise un modèle comparable en premier. La structure concurrentielle du secteur de l’IA signifie que la retenue d’un acteur n’empêche pas la capacité sous-jacente de finir par se répandre. Les cadres de gouvernance volontaires fonctionnent mieux lorsqu’ils génèrent des normes partagées à l’échelle d’un secteur. Lorsque le secteur est structuré autour d’une concurrence intense entre un petit nombre d’organisations, la retenue volontaire d’un seul acteur ne résout pas la question plus large de l’accès.

Les analystes notent que ce que Mythos fait aujourd’hui dans un environnement restreint, les modèles accessibles au public sont susceptibles de le reproduire d’ici une à deux générations de modèles. La prochaine phase de la loi européenne sur l’IA entrera en vigueur en août 2026, introduisant des pistes d’audit automatisées, des exigences de cybersécurité pour les systèmes d’IA classés à haut risque, des obligations de signalement des incidents et des sanctions pouvant atteindre 3 % du chiffre d’affaires mondial. Le cadre de l’UE marque un tournant vers une gouvernance contraignante, mais son champ d’application par rapport au rythme et à la répartition internationale du développement de l’IA de pointe reste à démontrer.

La marche à suivre

Anthropic reconnaît que des capacités telles que celles démontrées par Mythos se généraliseront au-delà des acteurs engagés à les déployer en toute sécurité, ce qui pourrait avoir des répercussions sur les économies, la sécurité publique et la sécurité nationale. La réponse de l’entreprise, considérée dans son ensemble, témoigne d’une volonté sérieuse de gérer ce risque dans le cadre de dispositifs volontaires et de la prise de décision privée. La politique de développement responsable, le projet Glasswing, les briefings proactifs auprès des gouvernements et la fiche technique détaillée constituent chacun des contributions substantielles. Ils sont également tous le fruit du jugement d’une seule entité privée, opérant sans obligation de rendre des comptes à des instances externes.

L’affaire Mythos n’appelle pas tant une réévaluation du comportement d’Anthropic qu’une vision lucide de ce que la gouvernance volontaire peut réellement soutenir à la frontière du développement de l’IA. Les gouvernements des deux côtés de l’Atlantique ont été briefés de manière informelle d’un modèle dont les capacités ont des conséquences pour les infrastructures critiques et la sécurité nationale. Aucune obligation de notification contraignante n’existait. Aucune autorité technique indépendante n’y avait eu accès au préalable. Aucun mécanisme de coordination internationale n’était en place.

Aucune organisation ne peut relever ces défis à elle seule. Les développeurs d’IA de pointe, les éditeurs de logiciels, les chercheurs en sécurité, les responsables de projets open source et les gouvernements ont tous un rôle essentiel à jouer. L’affaire Mythos a fait de cette observation non pas une simple déclaration d’intention, mais un problème politique qui nécessite des réponses institutionnelles concrètes. La question à laquelle sont désormais confrontés les décideurs politiques est de savoir si ces réponses prendront forme avant que le prochain seuil de capacité ne soit atteint.

Ce texte est une adaptation de l’article de Reyhan Damalan intitulé « Claude Mythos Preview établit une nouvelle référence en matière de capacités d’IA et soulève des questions de gouvernance ».

Le mois dernier à Genève

29e session de la CSTD

La 29e session de la Commission de la science et de la technologie au service du développement (CSTD) s’est tenue du 20 au 24 avril 2026 au Palais des Nations à Genève, en Suisse.

Pour sa 29e session, le programme a abordé le thème prioritaire « Science, technologie et innovation à l’ère de l’intelligence artificielle » et a entendu des présentations sur le rapport relatif aux activités de coopération technique dans les domaines de la science, de la technologie et de l’innovation.

Les membres de la CSTD ont également examiné les progrès réalisés dans la mise en œuvre et le suivi des résultats du Sommet mondial sur la société de l’information aux niveaux régional et international.

La session a également compris une séance d’information sur les projets du Groupe scientifique international sur l’IA, un rapport sur les étapes clés de la mise en œuvre du Pacte numérique mondial (GDC), ainsi qu’une occasion d’examiner le projet de feuille de route conjointe pour la cohérence entre le SMSI et le GDC.

Enfin, les membres de la CSTD ont adopté deux résolutions sur le SMSI et la science, la technologie et l’innovation au service du développement.

People, Person, Crowd, Indoors, Computer Hardware, Electronics, Hardware, Monitor, Screen, Cinema, Theater, Concert, Audience, Tony Meléndez, Abimael Guzmán, Stephen K. Amos, Li Hongli, Josh Grelle, Weird MC, Ole Anderson, La Parka

Source de l’image : publication Innovation X de la CNUCED

Élaboration de la stratégie suisse pour le Sommet sur l’IA

Un rapport destiné à éclairer la planification stratégique du Sommet sur l’IA de Genève 2027, synthétisant les contributions d’une table ronde multipartite et de plus de 50 contributions écrites afin d’influencer la stratégie de la Suisse pour l’organisation du Sommet sur l’IA, a été rendu public.

La principale conclusion de « Shaping Switzerland’s AI Summit Strategy » est que l’avantage comparatif de la Suisse ne réside pas dans l’envergure technologique, mais dans sa capacité à rassembler en toute confiance, sa gouvernance pragmatique et sa crédibilité institutionnelle. Sa neutralité, ses institutions solides, sa base de recherche (par exemple l’ETH/l’EPFL) et l’écosystème multilatéral de Genève la positionnent comme un facilitateur de la coopération pratique et intersectorielle. Cependant, des lacunes subsistent en matière d’investissement et de mise à l’échelle des innovations sur le marché.

Deux grands axes prioritaires se dégagent. Premièrement, une infrastructure d’IA fiable et souveraine, comprenant des modèles ouverts, l’interopérabilité et la réduction de la dépendance vis-à-vis des fournisseurs dominants – parallèlement à un déficit notoire de la Suisse en matière d’accès à des capacités de calcul IA de niveau industriel. Deuxièmement, l’impact de l’IA sur les droits de l’Homme, la sécurité et le droit humanitaire, en particulier en ce qui concerne l’utilisation militaire, la surveillance et la préservation de l’autonomie humaine. Les préoccupations transversales incluent la culture de l’IA, l’adoption par les PME, la préparation du secteur public et l’accès équitable pour les pays en développement.

Sur le plan stratégique, Genève 2027 devrait être conçue comme une plateforme de mise en œuvre, ont souligné les contributeurs, fournissant un ensemble limité d’outils pratiques et réutilisables à l’échelle internationale, soutenus par un processus préparatoire inclusif et des mécanismes de suivi.

Semaine de la cybersécurité de Genève 2026

L’Institut des Nations Unies pour la recherche sur le désarmement (UNIDIR) et le Département fédéral des affaires étrangères (DFAE) co-organisent la semaine de la cybersécurité à Genève du 4 au 8 mai 2026, réunissant des décideurs politiques, des diplomates, des experts techniques, des leaders du secteur privé, des universitaires et des représentants de la société civile dans divers lieux à Genève et en ligne pour une semaine de discussions sur la stabilité cybernétique, la résilience, la gouvernance, la numérisation et les implications sécuritaires des technologies émergentes, y compris l’IA.De retour après sa première édition, cet événement se positionne comme une réponse à un environnement cybernétique et géopolitique plus fragile. Organisée sous le thème « Faire progresser la coopération mondiale dans le cyberespace », la semaine de la cybersécurité à Genève 2026 intervient à un moment où l’insécurité cybernétique s’accroît, où les tensions géopolitiques s’intensifient et où les changements technologiques s’accélèrent. Le programme comprendra près de 90 événements et renforcera le rôle de Genève en tant que centre de cyberdiplomatie, de coopération internationale et de gouvernance numérique.

Opportunité : Devenez Chargé de recherche

Diplo a le plaisir de lancer un nouvel appel à candidatures pour les boursiers « Digital Watch » (2026).

Qu’est-ce que le Digital Watch Observatory ?

Le Digital Watch Observatory(DW) est un centre d’observation complet et une source d’information unique sur la gouvernance numérique. Il suit les dernières évolutions, fournit des aperçus et des analyses des politiques, et rassemble des informations sur les thèmes clés, les technologies, les processus, les acteurs politiques, les événements et les ressources.

Le DW s’adresse aux diplomates, aux décideurs politiques, aux chercheurs, aux acteurs de la société civile, aux représentants du monde des affaires et à d’autres parties prenantes qui ont besoin d’informations fiables, structurées, impartiales et actualisées sur les questions de gouvernance numérique.

Son contenu s’articule autour de :

Des thèmes, allant de la cybercriminalité et de la liberté d’expression à la gouvernance des données et aux infrastructures critiques.
Des technologies telles que l’intelligence artificielle, l’informatique quantique et les semi-conducteurs.
Des processus, notamment le Mécanisme mondial des Nations unies sur la sécurité des TIC, le Forum sur la gouvernance de l’Internet, le processus du Pacte numérique mondial, et bien d’autres encore.
Les acteurs politiques, tels que les pays, les entités techniques, les associations professionnelles, les entités des Nations unies et d’autres organisations internationales et régionales.
Des ressources, notamment des conventions, des résolutions, des lois et règlements, des rapports, etc.
Événements, tels que réunions, négociations, conférences et consultations.

Cette structure est complétée par :

Des mises à jour quotidiennes, des analyses régulières et des bulletins d’information hebdomadaires et mensuels qui suivent et expliquent les développements les plus pertinents dans le domaine de la gouvernance numérique.

Quel est le rôle d’un chargé de recherches ?

Les chargés de recherches (KF) jouent un rôle central dans la capacité de ce centre d’observation à fournir une couverture complète, précise et actualisée de domaines spécifiques de la gouvernance numérique.

Chaque chargé de recherches est censé couvrir un ou plusieurs domaines d’expertise et contribuer à garantir que le DW reste précis, pertinent, complet et impartial. Cela signifie :

Suivre et analyser les évolutions liées au(x) domaine(s) d’expertise attribué(s) et veiller à ce qu’elles soient prises en compte dans les mises à jour quotidiennes et les analyses régulières.
Veiller à ce que les pages du DW attribuées soient exactes, à jour et solides sur le fond.
Suivre les événements pertinents pour son ou ses domaines d’expertise et contribuer à ce que les réunions, négociations et discussions importantes soient reflétées sur le DW.
Identifier les ressources clés pertinentes pour son ou ses domaines d’expertise, telles que les résolutions des Nations unies et autres documents adoptés au niveau intergouvernemental, les lois, les réglementations, les rapports et les documents d’orientation.

Contribuer à une meilleure couverture des organisations, des pays et des autres acteurs clés de la gouvernance numérique.

Contribuer, le cas échéant, à des bulletins d’information, des documents de politique et de recherche, ainsi qu’à d’autres produits de connaissance.

Les chargés de recherches peuvent également avoir l’occasion de contribuer à l’écosystème de connaissances plus large de Diplo, notamment aux cours, aux discussions et aux initiatives thématiques.

Qui peut postuler ?

À une époque où l’espace public regorge de contenus générés par l’IA, nous recherchons plus qu’une simple personne capable d’utiliser l’IA pour résumer des actualités ou réécrire des ressources en ligne.

Les KF auront accès à des outils d’IA sur mesure pour les aider dans leur travail, mais le poste exige une expertise dans le domaine, un esprit critique et la capacité d’identifier ce qui est important, ce qui manque et ce qui mérite une analyse plus approfondie.

Plus précisément, nous recherchons des candidats qui :

Possèdent une solide expertise en gouvernance numérique, fondée sur une expérience professionnelle, des recherches universitaires, un engagement politique ou une combinaison de ces éléments.
Souhaitent continuer à développer cette expertise.
Savent où chercher et quoi rechercher afin d’assurer une couverture exhaustive des sujets, technologies, processus, etc. qui leur sont assignés.
Sont capables d’identifier les développements majeurs, les controverses politiques, les débats clés et les tendances émergentes dans le paysage de la gouvernance numérique, et de les couvrir avec précision et impartialité.

Cela implique de combiner une expertise thématique avec un jugement éditorial, une connaissance des politiques et un sens aigu de la curation des connaissances.

Les candidats doivent également disposer :

Être disponible pour contribuer régulièrement. Le programme se déroule en ligne et nécessite un engagement d’au moins 8 heures par semaine.
De solides compétences analytiques et rédactionnelles en anglais.
Des compétences de base dans l’utilisation du web et des réseaux sociaux, ainsi qu’une bonne connaissance des outils d’IA générative.

Ce que nous offrons

Les boursiers Digital Watch Knowledge bénéficieront :

d’une formation initiale et des conseils sur l’approche éditoriale et de curation de Digital Watch.
d’une formation aux processus de travail du Digital Watch et aux outils numériques/d’IA.
d’une rémunération.
d’une visibilité de leur travail auprès des utilisateurs de DW (communautés diplomatiques à Genève et dans d’autres centres diplomatiques, professionnels de tous les groupes de parties prenantes traitant de sujets numériques, etc.)
d’opportunités de promouvoir leurs recherches sur la gouvernance numérique via les réseaux de DW et Diplo.
de l’adhésion à une communauté mondiale d’experts et de professionnels travaillant sur la gouvernance numérique.

Les boursiers sont engagés à titre de consultants/rémunérés ; ce poste ne constitue pas un emploi au sein de la DiploFoundation.

Comment postuler

Les candidats intéressés sont invités à remplir le formulaire de candidature.
Date limite de candidature : 31 mai 2026

Weekly #262 The age of tech exceptionalism is ending

8 – 15 May 2026

HIGHLIGHT OF THE WEEK

The age of tech exceptionalism is ending

For years, tech companies defended themselves with the same argument: platforms are neutral. They host content, process information, and connect users, but they are not responsible for what people do with those systems. That distinction is now rapidly collapsing.

This week offered a glimpse of what comes next.

Families in the USA have launched multiple lawsuits against OpenAI, alleging that ChatGPT interactions contributed to violent acts or dangerous behaviour. One case linked the chatbot to planning surrounding a 2025 shooting at Florida State University.The lawsuit contends that OpenAI failed to identify escalating risk indicators within the conversations and did not adequately prevent harmful guidance. It argues the system ‘failed to connect the dots’ despite the alleged shooter’s repeated questions about suicide, terrorism and mass shootings.

A California lawsuit claimed ChatGPT provided advice about combining medication resulting in a fatal overdose of 19-year-old Sam Nelson. His parents have now filed a wrongful death lawsuit against OpenAI, are seeking financial damages, as well as a pause to ChatGPT Health, OpenAI’s recently launched feature designed to offer medical help.

Researchers reported this week that 10 out of 10 investigated chatbots provide users advice on undertaking school shootings, including ChatGPT, Gemini, Claude, Copilot, Meta AI, DeepSeek, Perplexity, MyAI, Character.ai and Replika.

The details differ, but the underlying concern is the same: AI systems are increasingly interacting with users in ways that resemble guidance, persuasion, or emotional engagement – all very human things to do. And the more successful those systems become at simulating human interaction, the harder it becomes for companies to argue they are merely neutral intermediaries.

That tension is also surfacing inside OpenAI itself. The ongoing legal battle between Elon Musk and Sam Altman is often framed as a personality conflict, but beneath it lies a deeper dispute over what AI companies are becoming. Musk argues OpenAI abandoned its nonprofit mission and transformed into a commercial empire chasing market dominance. Altman, meanwhile, defends the shift as necessary to survive the escalating competition for compute, talent, and investment.

Jury, Person, Adult, Male, Man, Indoors, People, Face, Head, Publication, Book, Courtroom, Room, Crowd

Multiple legal disputes highlight escalating regulatory and litigation pressure on major technology companies across AI, media, retail, and consumer data use. The EU Court of Justice upheld an Italian order requiring Meta to compensate publishers for news snippets, intensifying copyright disputes over content reuse and AI training data. In Milan, families and MOIGE have filed a case against Meta and TikTok over alleged failures to protect minors on social media platforms. France’s top court rejected Amazon’s challenge to a minimum book delivery fee designed to protect independent bookstores against global platform dominance. Texas has sued Netflix for allegedly collecting user data without consent, including viewing habits and engagement tactics such as autoplay, under deceptive trade practices law.

The big picture. Regulators struggled to apply existing legal frameworks to platforms that claimed neutrality or technological exceptionalism. That position is becoming harder to sustain.

IN OTHER NEWS LAST WEEK

AI governance

The UN. The UN has invited member states and stakeholders to nominate co-chairs for the thematic discussions and participate in the first Global Dialogue on AI Governance in Geneva, scheduled for 6–7 July 2026. The discussions will be organised around four themes, each of which will be jointly chaired by one member state and one stakeholder representative, with the aim of fostering multistakeholder exchanges on experiences, best practices, and policy cooperation. Governments are asked to nominate high-level representatives, while stakeholders are encouraged to nominate senior experts relevant to the selected theme.

The EU. The Council of the European Union has confirmed agreement on a compromise text for the Digital Omnibus on AI, a proposal intended to simplify parts of the EU AI Act’s implementation. The Permanent Representatives Committee confirmed the agreement on 13 May 2026, following informal negotiations between the EU institutions on 6 May. The Council Presidency was authorised to send a letter to the European Parliament stating that, if Parliament adopts the text at first reading, the Council will approve Parliament’s position.

The European Commission has updated the ERA Living Guidelines on the responsible use of generative AI in research. New recommendations address risks linked to the use of generative AI by third parties, including in meetings, note-taking, summaries and document overviews, where confidential information, data protection or intellectual property rights may be affected. The guidelines encourage researchers and organisations to inform third parties about the use of such tools and related risks. A specific addition concerns the risk of ‘hidden prompts’, where instructions may be secretly embedded in documents or inputs to influence generative AI tools. The guidelines call on research funding organisations to remain aware of such risks, set rules prohibiting manipulation where relevant, and introduce appropriate safeguards in IT systems used to process information.

Australia. Australia’s National AI Centre has launched AI.gov.au, a national platform designed to help organisations adopt AI safely and responsibly in line with the National AI Plan. The platform provides a central source of guidance, tools and resources to support businesses and not-for-profits. It aims to help users identify AI opportunities, plan implementation, manage risks and build internal capability.

South Korea. The Office of National Security of South Korea held a meeting to review how government agencies are responding to AI-driven cyber threats. The session focused on the growing risks posed by the misuse of advanced AI technologies.Officials from multiple ministries attended, including science, defence and intelligence bodies, to coordinate responses. Discussions also covered practical measures to support rapid responses to cybersecurity vulnerabilities across public and private sectors.

The Trump-Xi summit

Trade and technology were among the main topics expected to dominate the Trump-Xi summit held on Thursday and Friday in Beijing, the first state visit of a US President to China in 8 years.

A litany of CEOs accompanied Trump on the trip, including Tesla’s Elon Musk, Nvidia’s Jensen Huang, Apple’s Tim Cook, GE Aerospace’s Larry Culp, Boeing’s Kelly Ortberg, Meta’s Dina Powell McCormick, BlackRock’s Larry Fink, Blackstone’s Stephen Schwarzman, Micron’s Sanjay Mehrotra, Mastercard’s Michael Miebach, Qualcomm’s Cristiano Amon and Visa’s Ryan McInerney.

But as much as the public is assured that progress was made, details remain scarce.

When it comes to tech matters, AI was expected to be a central topic of discussion, with reports in the lead up to the summit suggesting that the USA and China are considering launching formal discussions on AI. ‘The two AI superpowers are going to start talking. We’re going to set up a protocol in terms of how do we go forward with best practices for AI to make sure nonstate actors don’t get a hold of these models,’ US Treasury Secretary Scott Bessent told CNBC. Bessent also added ‘The reason we are able to have wholesome discussions with the Chinese on AI is because we are in the lead,’ and candidly admitted: ‘I do not think we would be having the same discussions if they were this far ahead of us.’

Part of the AI story are, of course, the chips needed for AI. The USA has tried to limit China’s AI development by restricting sales of advanced chips, with Jensen Huang’s Nvidia and Huang himself caught in the middle. Yet, reports suggest that the USA has cleared around 10 Chinese firms to buy the H200, Nvidia’s second-most powerful AI chip, the H200. However, no deliveries were made. This is likely the breakthrough Huang was seeking in China.

And part of the chips story are the minerals the chips are made of, and on those China has an upper hand which is deftly uses as a lever in its trade relations with the USA. US Trade Representative Jamieson Greer told Bloomberg that rare earth exports from China are improving, although ‌Beijing is still slow to approve some export licences.

However, Greer also stated that chip export controls were not a major part of the talks.

It remains to be seen whether more outcomes will be shared in the following days.

Council of the EU extends cyber sanctions framework until 2027

The Council of the European Union has extended restrictive measures against individuals and entities involved in cyber-attacks threatening the EU and its member states until 18 May 2027. The legal framework behind the sanctions regime had already been extended until 18 May 2028.

The framework allows the EU to impose targeted sanctions on persons or entities involved in significant cyber-attacks that constitute an external threat to the Union or its member states. Measures can also be imposed in response to cyber-attacks against third countries or international organisations, where they support Common Foreign and Security Policy objectives.

Why does it matter? By extending cyber sanctions listings, the EU is reinforcing its use of diplomatic and economic measures to deter malicious cyber activity, attribute responsibility and signal that significant cyber-attacks can carry geopolitical consequences.

Thales confirms data breach

French defence and cybersecurity giant Thales Group has confirmed a third-party data breach after a dataset allegedly linked to the company appeared on a cybercrime forum. The leaked data appears to be linked to LuxTrust, a Luxembourg-based digital identity provider that uses Thales services.

Thales said the breach originated from an outsourced service and stated that there was no operational impact on the company or its customers based on current information. The company has notified France’s data protection authority, CNIL, and launched an investigation.

Why does it matter? The exposed sample includes names, email addresses, account metadata, and company-related fields, suggesting the information likely originated from an outsourced or partner platform rather than LuxTrust’s internal systems. However, even limited identity data could enable phishing and social engineering attacks, particularly because LuxTrust supports authentication services used by governments, financial institutions, and enterprises.

Pakistan enacts crypto law

Pakistan has enacted the Virtual Assets Act 2026, establishing a permanent legal foundation for the Pakistan Virtual Assets Regulatory Authority. The move formalises oversight of the country’s growing digital asset sector, which had previously operated under temporary rules introduced in 2025.

Under the new framework, PVARA is responsible for licensing, regulating and supervising virtual assets and virtual asset service providers operating in Pakistan. The law gives the authority powers to issue, suspend and revoke licences, while unlicensed operations can face fines and criminal penalties.

PVARA is working with national institutions as it transitions into an operational regulator, including through a regulatory sandbox and rules requiring prior authorisation for virtual asset pilots, partnerships and implementations involving users in Pakistan.

The legislation comes as Pakistan seeks to attract regulated digital asset activity while tightening oversight of financial crime risks. Officials have also linked virtual asset regulation with broader ambitions in digital finance, AI-powered payments and workforce preparation for emerging technologies.

US CLARITY Act moves forward amid split over crypto market rules

The United States Senate Banking, Housing, and Urban Affairs Committee has advanced the Digital Asset Market Clarity Act in a 15–9 vote, marking another step towards establishing a federal framework for digital asset markets.

A central issue in the revised text is how to regulate stablecoin-related activity. The bill seeks to prevent stablecoins from functioning like bank deposits by limiting passive yield on customer holdings, while still allowing certain rewards linked to user activity or platform use.

The bill also continues debate over decentralised finance, including how far regulation should extend to developers, protocols and infrastructure providers that do not directly custody user funds.

Ethics provisions were among the most contested issues during the markup process, with lawmakers divided over whether and how to restrict potential conflicts of interest involving public officials and cryptoasset activities.

Further hurdles remain before the legislation can become law. The bill will need to advance through the full Senate, be reconciled with other Senate work on digital asset regulation and secure agreement with the House of Representatives before reaching the President’s desk.

View all updates

LAST WEEK IN GENEVA

Intergovernmental Group of Experts on E-commerce and the Digital Economy 9th session

The UNCTAD Intergovernmental Group of Experts on E-commerce and the Digital Economy met in Geneva for its 9th session from 11–13 May 2026 to address how digital trade affects fiscal systems in developing countries. The agenda focused on strengthening tax revenues in the context of e-commerce, including VAT and customs challenges, platform-based taxation, and the implications of cross-border digital transactions. Sessions explored revenue risks, administrative digital tools, and the role of platforms as tax intermediaries, alongside international and regional cooperation to improve tax readiness. The meeting also covered measuring the digital economy and advancing capacity-building to support inclusive and sustainable digital development strategies.

UN Virtual Worlds Day 2026
The 3rd UN Virtual Worlds Day, co-organised by the International Telecommunication Union (ITU) and multiple UN agencies and partners, took place in Geneva on 11–12 May 2026. It convened ambassadors, ministers, city leaders, UN entities, industry pioneers and innovators for two days of high-level dialogue on AI-driven cities, immersive environments and spatial intelligence. Discussions focused on how frontier technologies are reshaping urban governance and public services, including the emerging citiverse concept and its link to the Global Digital Compact. It concluded with executive briefing on AI, spatial intelligence and AI-enabled citiverse, and the Call to Action for inclusive digital futures.

READING CORNER

Executive Briefing: AI, Spatial Intelligence and the AI Enabled Citiverse

This executive briefing has been developed by the 20 UN entities co-organising UN Virtual Worlds Day 2026. The briefing assesses how AI, spatial intelligence and the AI-enabled citiverse are transforming urban governance, urban development and global digital cooperation.

OPPORTUNITY

Opportunity: Become a Knowledge Fellow

Weekly #261 Pulling at the threads of AI governance

1 – 8 May 2026

HIGHLIGHT OF THE WEEK

Pulling at the threads of AI governance

AI governance increasingly resembles a tangled ball of yarn: regulation, cybersecurity, infrastructure, labour markets, semiconductors, and geopolitics pulling on one another simultaneously.

There is an older word for such a ball of thread: clew. Historically, a clew was not just yarn, but a guide through a maze — the thread used in mythology to navigate complexity and find a way out. It is also the root of the modern word clue.

This week in AI governance felt like standing at the entrance to a maze with a clew in hand, faced with multiple threads.

Rope, Adult, Male, Man, Person, Face, Head

Europe’s AI Act. Europe’s AI rulebook edged forward this week after negotiators reached a provisional agreement on the latest phase of the EU AI Act omnibus discussions, which aims to simplify parts of the Union’s digital rulebook and ease implementation burdens. The provisional agreement sets new application dates of 2 December 2027 for stand-alone high-risk AI systems and 2 August 2028 for high-risk AI systems embedded in products.

The agreement also extends certain simplification measures beyond SMEs to small mid-caps, while keeping some safeguards. The agreement also reinforces the AI Office’s powers. The text must still be endorsed by both the Council and the European Parliament before formal adoption.

National AI strategies. National AI strategies are also becoming more assertive — and more tied to economic sovereignty. Papua New Guinea has outlined a national approach to AI focused on data sovereignty, trusted public infrastructure, and new legislation, underpinned by four elements of the framework: Strengthening existing digital foundations such as SevisPass and SevisDEx, establishing a National AI Register, adopting sovereign data governance, introducing new laws, including a National Artificial Intelligence Act and a Data Governance and Protection Act. Kazakhstan reviewed proposals to expand AI deployment across all sectors as part of its digital transformation agenda. Canada moved to strengthen domestic photonic semiconductor and AI capabilities by spinning off the National Research Council of Canada (NRC)’s Canadian Photonics Fabrication Centre (CPFC) into a commercial entity. The UAE launched a national AI security laboratory focused on certification and cyber resilience.

These initiatives vary widely in ambition and capacity, but they share a common premise: AI infrastructure, data, and chips are now viewed as strategic assets.

AI diplomacy. At the same time, AI diplomacy is accelerating. Australia and Japan expanded cooperation on economic security and critical technologies. The EU and Japan advanced joint work on AI governance and cross-border data flows. South Korea and the Netherlands discussed semiconductor and AI cooperation. India and France have discussed expanding cooperation in space, AI, applied mathematics and advanced technologies. Norway joined the Pax Silica initiative, which focuses on securing semiconductor supply chains.

Even geopolitical rivals appear to be cautiously reopening channels on AI governance. According to reports, the USA and China are considering launching formal discussions on AI, signalling recognition that some degree of coordination may become unavoidable as frontier systems grow more capable and globally consequential.

Cybersecurity. The security picture darkens as multiple warnings are issued in one week.

The UK’s National Cyber Security Centre warned that AI systems could dramatically accelerate the discovery and exploitation of software vulnerabilities, compressing the time between disclosure and attack. Separate guidance from the NCSC examined the growing risks posed by adversarial machine learning attacks, including model manipulation, prompt injection, and data poisoning techniques designed to undermine AI systems themselves.

Swisscom similarly warned that AI and geopolitical tensions are reshaping the cyber threat landscape, with automation, influence operations, and AI-enhanced cyber capabilities becoming increasingly intertwined.

The Australian Securities and Investments Commission (ASIC) has urged regulated entities to strengthen cyber resilience, warning that frontier AI could intensify cyber risks by exposing vulnerabilities at greater speed, scale and sophistication. ASIC said licensees and market participants should act now to improve their cybersecurity fundamentals rather than wait as advanced AI tools reshape the threat environment.

In the USA, Microsoft, Google, and xAI agreed to provide advanced AI models for government-led security stress testing. The initiative is designed to support pre-deployment evaluations and targeted research intended to improve understanding of frontier AI capabilities and their national security implications.

Other threads. Canada found OpenAI non-compliant in a privacy probe — scraping public data for training was overbroad and lacked consent. Meta fought an EU order in a closed hearing, seeking to avoid letting rival AI chatbots onto WhatsApp for free.

The user’s POV. For compliance teams, the EU deal offers breathing room, but not a free pass — the transparency deadlines actually tightened. For security professionals, the NCSC warnings are a call to audit ML pipelines now, not later. For everyone watching geopolitics, the US-China AI talks would be the first real signal that both capitals see cooperation as necessary. The question is whether they can agree on anything beyond the need to talk.

The bottom line. The challenge for governments, companies, and users alike is no longer simply building AI systems. It is learning which strands matter, which knots are tightening, and which clews still lead out of the maze.

IN OTHER NEWS LAST WEEK

Meta on trial(s)

Meta Platforms is facing growing legal and regulatory pressure both in the USA and Europe over claims that its social media platforms contribute to youth addiction and mental health problems.

In New Mexico, the state is seeking $3.7 billion and asking the court to declare Meta a public nuisance. The lawsuit alleges that Facebook, Instagram, and WhatsApp were designed in ways that encourage addictive behaviour among minors. It also claims that these platforms failed to adequately protect young users from harmful content and exploitation. The state is requesting major changes to the platforms, including age verification and restrictions on features such as autoplay and infinite scroll for minors. Meta claims the case concerns individual users, rather than the harm to the public as a whole.

Meta is also attempting to overturn a California jury verdict that found the company negligent in the design of its platforms and awarded damages to a young plaintiff who claimed that social media use contributed to her depression. Meta argues that the claims are barred by Section 230 of the Communications Decency Act and that the alleged harms were connected to online content rather than the platforms’ design features.

Why does it matter? Both cases are considered important because they may influence many similar lawsuits currently pending against social media companies.

Dutch court backs DigiD contract despite US data access fears

The District Court of The Hague has rejected an attempt by three Dutch citizens to block the government from renewing its contract with Solvinity, the company responsible for hosting and technically managing systems linked to DigiD.

The plaintiffs argued that Solvinity’s planned acquisition by US-based IT provider Kyndryl could place sensitive data from more than 16 million DigiD users under US jurisdiction, potentially exposing it to US authorities and creating risks to critical public services such as healthcare, pensions, taxes, and unemployment systems.

Despite these concerns, the court ruled in favour of the Dutch State, allowing the agreement to proceed. Judges did not accept arguments that the deal would immediately threaten data security or justify halting the contract.

What’s next? The decision leaves further scrutiny to the Investment Assessment Office, which is reviewing national security risks linked to the acquisition.

Why does it matter? The case highlights ongoing tensions around digital sovereignty and data protection in the Netherlands.

End-to-end encrypted messaging on Instagram ends as wider encryption battles grow

As of 8 May, end-to-end encrypted messaging on Instagram is officially over. Meta has switched off the feature globally, abandoning plans to expand the privacy technology across the platform after years of promoting encrypted communication as the future of messaging.

At the same time, Apple and Meta are opposing Canada’s proposed Bill C-22, which they say could force companies to weaken encryption or build government-access mechanisms into their products. Canadian authorities argue the bill would help law enforcement respond more quickly to security threats.

Why does it matter? End-to-end encryption is widely seen as a core privacy protection because it limits access to message content, including by the platform itself. This week’s developments underline the questions about how major platforms prioritise privacy features, user safety, product complexity and interoperability across their messaging services.

View all updates

LAST WEEK IN GENEVA

WTO resumes talks as 19 members back e-commerce moratorium pledge

The WTO’s General Council has met in Geneva for the first time since the 14th Ministerial Conference (MC14), after negotiators in Yaoundé narrowly missed agreements on several major files, including the future of the long-running e-commerce moratorium and broader WTO reform.

The newly elected chair, Ambassador Clare Kelly, said members remained committed to preserving the careful balance reached during negotiations in Cameroon and avoiding a return to earlier positions.

The discussions follow the expiry on 31 March of the WTO moratorium on customs duties on electronic transmissions, a temporary arrangement first adopted in 1998 that prevented countries from imposing tariffs on digital trade flows such as software downloads, streaming services, and other online transmissions. Ministers at MC14 failed to agree on another extension, exposing deep divisions over how long the moratorium should continue and how governments should respond to rapid technological developments such as AI and 3D printing.

Despite the lapse, a group of 19 WTO members announced they would continue not imposing customs duties on electronic transmissions among themselves. In a joint statement circulated at the WTO, the group said the arrangement would provide predictability and certainty for businesses and consumers while multilateral negotiations continue. The group includes Argentina, Australia, Costa Rica, Ecuador, Guatemala, Iceland, Israel, Japan, South Korea, Malaysia, Mexico, New Zealand, Norway, Panama, Paraguay, Singapore, Separate Customs Territory of Taiwan, Penghu, Kinmen and Matsu, the USA, and Uruguay.

Türkiye also signalled new flexibility during the General Council meeting, announcing it would not block consensus on a temporary extension of the moratorium. However, Brazil maintained its opposition to a four-year extension of the moratorium.

What’s next? The chair announced further consultations on e-commerce and WTO reform, with plans to report back to members in July.

Geneva Cyber Week 2026

The UN Institute for Disarmament Research (UNDIR) and the Swiss Federal Department of Foreign Affairs (FDFA) are co-hosting Geneva Cyber Week from 4 to 8 May 2026, bringing policymakers, diplomats, technical experts, industry leaders, academics, and civil society representatives to venues across Geneva and online for a week of discussions on cyber stability, resilience, governance, digitalisation, and the security implications of emerging technologies, including AI.

Returning after its inaugural edition, the event is being positioned as a response to a more fragile cyber and geopolitical environment. Held under the theme ‘Advancing Global Cooperation in Cyberspace’, Geneva Cyber Week 2026 comes at a moment of mounting cyber insecurity, intensifying geopolitical tension, and rapid technological change. The programme features nearly 90 events and reinforced Geneva’s role as a centre for cyber diplomacy, international cooperation, and digital governance.

As part of Geneva Cyber Week, UNIDIR organised the Cyber Stability Conference 2026, on 4–5 May in Geneva and online, bringing together governments, international organisations, industry, academia, and civil society to discuss ICT security and cyber governance. Under the theme ‘Cyber governance in an era of technological revolution: Past lessons, present realities and future frontiers,’ discussions explored how international cyber stability frameworks are adapting to rapid technological change, including AI and quantum computing, while reflecting on lessons from past cyber diplomacy processes and current security challenges.

Multi-year expert meeting on investment, innovation and entrepreneurship for productive capacity-building and sustainable development, 12th session

UNCTAD’s Multi-year Expert Meeting on Investment, Innovation and Entrepreneurship for Productive Capacity-building and Sustainable Development met for its twelfth session 4-5 May. The experts warned that AI and other strategic technologies are reshaping global investment patterns, concentrating capital in a handful of sectors and countries while leaving many developing economies behind. Discussions at the meeting focused on how developing countries can compete in AI-related sectors, strengthen domestic innovation ecosystems and ensure that AI-driven investment translates into broader development gains.

OSCE Conference ‘Anticipating technologies – for a safe and humane future’ opens in Geneva

The Swiss Chairpersonship of the Organization for Security and Co-operation in Europe opened a two-day high-level conference on anticipatory technologies in Geneva on 7 May. The event is examining how foresight, dialogue, and international cooperation can help reduce misunderstandings, build trust, and strengthen security across the OSCE region amid rapid technological change.

The programme includes discussions on anticipating technological change and its geopolitical impact, water and energy security in the digital age, and the role of AI in early warning and conflict prevention.

The conference also highlights Geneva’s role as a meeting point for science and diplomacy, including through institutions such as CERN, the Geneva Science and Diplomacy Anticipator, and the Open Quantum Institute.

The event forms part of the Chairpersonship’s priority to connect scientific and technological anticipation with policy action.

READING CORNER

UNESCO report warns over global quantum research inequality

Quantum research access, UNESCO says, remains concentrated in wealthier economies.

UNDP highlights challenges in public sector digital transformation outcomes

Public sector transformation requires more than technology, demanding systemic reform in governance, procurement, and talent to meet citizens’ real needs.

learning education ideas insight intelligence study concept 1

ILO warns lifelong learning is critical for the future AI economy

Global labour markets face rapid transformation as the ILO highlights growing AI and skills challenges.

OPPORTUNITY

Opportunity: Become a Knowledge Fellow

Opportunity: Become a Knowledge Fellow

DiploFoundation is pleased to launch a new call for applications for Digital Watch Knowledge Fellows (2026).

What is the Digital Watch Observatory?

The Digital Watch Observatory (DW) is a comprehensive observatory and one-stop shop source of information on digital governance. It tracks latest developments, provides policy overviews and analysis, and curates information on key topics, technologies, processes, policy players, events, and resources.

DW is designed for diplomats, policymakers, researchers, civil society actors, business representatives, and other stakeholders who need reliable, structured, impartial, and up-to-date information on digital governance issues.

Its content is organised around:

Topics, from cybercrime and freedom of expression, to data governance and critical infrastructure.
Technologies such as artificial intelligence, quantum computing, and semiconductors.
Processes including the UN Global Mechanism on ICT Security, the Internet Governance Forum, the Global Digital Compact process, and more.
Policy players such as countries, technical entities, business associations, UN entities, and other international and regional organisations.
Resources, including conventions, resolutions, laws and regulations, reports, and more.
Events, such as meetings, negotiations, conferences, and consultations.

This structure is complemented by:

Daily updates, regular analyses, and weekly and monthly newsletters that track and explain the most relevant developments across the digital governance landscape.

What is the role of a Knowledge Fellow?

Knowledge Fellows (KF) are central to the observatory’s ability to provide comprehensive, accurate, and up-to-date coverage of specific areas of digital governance.

Each KF is expected to cover one or more areas of expertise and help ensure that DW remains accurate, relevant, and complete, and impartial. This means:

Monitoring and analysing developments related to the assigned area(s) of expertise and ensuring these are reflected in daily updates and regular analyses.
Keeping assigned DW pages accurate, up-to-date, and substantively strong.
Tracking events relevant to their area(s) of expertise and helping ensure that important meetings, negotiations and discussions are reflected in DW.
Identifying key resources relevant to their area(s) of expertise such as UN resolutions and other intergovernmentally agreed documents, laws, regulations, reports, and policy papers.
Supporting stronger coverage of organisations, countries, and other key actors in digital governance.
Contributing, when relevant, to newsletters, policy and research papers, and other knowledge products.

Knowledge Fellows may also have opportunities to contribute to Diplo’s wider knowledge ecology, including courses, discussions, and thematic initiatives.

Who should apply?

At a time when the public space is abundant with AI-generated content, we are looking for more than just someone who can use AI to summarise news or rewrite online resources.

KF will have access to custom-made AI tools to support them in their work, but the role requires subject expertise, critical judgement, and the ability to identify what is important, what is missing, and what deserves deeper analysis.

Specifically, we are looking for applicants who:

Have a strong expertise in digital governance, grounded in professional experience, academic research, policy engagement, or a combination of these.
Are interested in continuing to develop this expertise.
Know where to look and what to look for in order to ensure a comprehensive coverage of assigned topics, technologies, processes, etc.
Can identify major developments, policy controversies, key debates, and emerging trends in the digital governance landscape, and cover them accurately and impartially.

This means combining subject expertise with editorial judgement, policy awareness, and a strong sense of knowledge curation.

Applicants must also have:

Availability to contribute on a regular basis. The fellowship is conducted online, with an expected commitment of at least 8 hours per week.
Strong analytical and writing skills in English.
Basic skills in using web and social media, as well as familiarity with generative AI tools.

What we offer

Digital Watch Knowledge Fellows will benefit from:

Onboarding and guidance on Digital Watch’s editorial and curation approach.
Training on observatory workflows and digital/AI tools.
Remuneration.
Visibility for their work among DW users (diplomatic communities in Geneva and other diplomatic centres, professionals from across all stakeholder groups dealing with digital topics, etc.)
Opportunities to promote their digital governance-related research through DW and Diplo networks.
Membership in a global community of experts and professionals working on digital governance.

Fellows are engaged on a consultancy/fee basis; The role does not constitute employment with DiploFoundation.

How to apply

Interested applicants are invited to complete the application form below.

Application deadline: 31 May 2026

Questions: dwapplications@diplomacy.edu

Link to form

Digital Watch newsletter – Issue 109 – April 2026

April 2026 in retrospect

Dear readers,

Visionary or outlandish statements about the future are a feature of tech industry discourse. But the rapid acceleration of generative AI seems to have shortened the timeline for many of these claims. April brought another wave of high-profile predictions. While some might be tempted to dismiss them as mere hype, there’s a strong reason to assess them: the quiet danger of designing the future without meaningful public input.

South Africa unveiled its first draft national AI policy and was quickly forced to withdraw it after reviewers found a critical flaw: it was riddled with fake sources and non-existent citations, likely generated by AI. This incident is rather illustrative of a broader problem with AI-generated laws. In this issue of the newsletter, we examine how to prevent fake laws governing real life.

In earty April, Anthropic announced Claude Mythos Preview, its most capable AI model to date, alongside the explicit decision not to make it publicly available. We look at the model’s capabilities, the reason behind restricting access to the model, as well as the governance questions the model has brought up.

We invite interested readers to join our team of Knowledge Fellows. Knowledge Fellows are central to the observatory’s ability to provide comprehensive, accurate, and up-to-date coverage of specific areas of digital governance. More details on what we are looking for and what we offer in return are available in the newsletter.

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

Snapshot: The developments that made waves in April

Technologies

The EU and the USA have launched a coordinated framework to strengthen resilience in critical minerals supply chains, combining a strategic Memorandum of Understanding (MoU) with an Action Plan. partnership aims to secure diversified and sustainable supply chains through joint project development in the EU, US, and third countries, supported by coordinated investment tools, risk reduction mechanisms, and improved business linkages.

Canada and Finland have set out a new agenda for cooperation on sovereign technology and AI, positioning advanced digital capabilities as central to economic resilience, security, and strategic autonomy in a contested global environment. Announced after talks in Ottawa, the agenda spans AI adoption across government and industry, high-performance computing, telecommunications, AI gigafactories (including support for Nokia’s AI gigafactory), quantum research, critical minerals, and trusted supply chains. Both countries plan to deepen coordination on sovereign AI infrastructure, reduce technological dependencies, support small and medium-sized enterprises, and expand telecom opportunities through initiatives such as the Global Coalition on Telecommunications.

Canada is increasing support for its quantum research ecosystem through new funding announced by the Natural Sciences and Engineering Research Council of Canada, aiming to strengthen the country’s scientific capacity, innovation base, and long-term leadership in a strategically important field. The initiative will back researchers, projects, and cross-institutional collaboration, advancing both fundamental science and applied development while helping translate quantum research into practical technological progress.

The UK government has identified six frontier technologies – AI, cybersecurity, advanced connectivity, engineering biology, quantum technologies, and semiconductors – as the pillars of its 2025 Modern Industrial Strategy and Digital and Technologies Sector Plan, aiming to strengthen digital capability, economic growth, national resilience, and long-term competitiveness. The agenda prioritises investment in next-generation telecoms, including 5G and future 6G, alongside expanded compute capacity, supercomputing infrastructure, and workforce development to reinforce the UK’s position as a leading EU AI hub.

Australian researchers have used a Wikipedia-based AI model to identify 100 emerging technologies gaining momentum ahead of 2026, offering a data-driven alternative to traditional forecasting methods often shaped by expert judgement. Drawing on thousands of Wikipedia entries, the analysis mapped more than 23,000 technologies to produce the ‘Momentum 100’ list, led by reinforcement learning and followed by blockchain, 3D printing, soft robotics, augmented reality, and other fast-developing fields.

Infrastructure

European technology providers Cubbit, SUSE, Elemento, and StorPool Storage have launched a joint Disaster Recovery Pack to help organisations maintain data access and operational continuity during disruptions caused by external technology dependencies. Presented at the European Data Summit in Berlin, the solution combines storage, compute, orchestration, networking, identity, observability, and management into a single deployable cloud software stack designed to reduce fragmentation and simplify recovery planning. By enabling critical workloads to be transferred to European-based infrastructure with limited disruption, the initiative seeks to meet practical disaster recovery needs while supporting wider efforts to reduce reliance on non-European cloud providers.

A new report, citing research by the Brussels-based Future of Technology Institute, warns that most of the EU defence agencies remain heavily dependent on US cloud and technology providers, raising concerns over exposure to a potential ‘kill switch’ scenario in which critical services could be restricted or disabled during political or strategic tensions. Open contracting data reviewed by the institute suggests that 23 of 28 EU and UK countries rely on US firms, either directly or through EU suppliers, using American cloud infrastructure, with 16 countries classified as high risk, including Germany, Finland, Poland, Denmark, Estonia, and the UK. Google Cloud, Microsoft, and Oracle are described as dominant providers in sensitive defence systems, while Austria is presented as a lower-risk case due to apparent reliance on sovereign alternatives.

Panthalassa has raised $140 million in Series B funding, led by Peter Thiel, to develop offshore systems that harness ocean wave energy to power AI computing as demand for data centre capacity accelerates. The company plans to build wave-powered nodes that generate electricity at sea, run AI computing on board, and transmit data through low-Earth-orbit satellites, offering a potential response to land-based data centres’ growing constraints on power supply, cooling, and infrastructure.

Security

By pairing AI-driven discovery with industry coordination, $100 million in usage credits, and funding for open-source security, the project Glasswing brings together major technology, cybersecurity, finance, and open-source actors, including AWS, Apple, Google, Microsoft, NVIDIA, Cisco, CrowdStrike, JPMorgan Chase, and the Linux Foundation, in a coordinated effort to use advanced AI to defend critical software infrastructure. Led by Anthropic’s Claude Mythos Preview model, the initiative aims to detect complex vulnerabilities at scale, with early findings uncovering thousands of previously unknown flaws across operating systems, browsers, and core digital infrastructure, some of which had remained hidden for decades.

A joint CISA advisory warns that Iranian-affiliated cyber actors are targeting internet-facing programmable logic controllers across US critical infrastructure, including Rockwell Automation and Allen-Bradley CompactLogix and Micro850 devices used in government, water, energy, and industrial systems. Active since at least March 2026, the campaign has disrupted PLC functions, manipulated project files, and altered HMI and SCADA displays, causing operational and financial damage.

Canada has introduced Level 1 of the Canadian Program for Cyber Security Certification, setting a baseline of cyber requirements for suppliers working on defence contracts as cyber threats increasingly target contractors, sensitive data, and critical supply chains. Phased implementation will begin in summer 2026, with certification required at the contract award stage, giving industry time to adapt while strengthening procurement trust and operational readiness.

Europol’s 2026 Internet Organised Crime Threat Assessment warns that the EU’s cybercrime landscape is becoming more complex, industrialised, and difficult to disrupt as criminal networks exploit encryption, proxies, fragmented online spaces, and AI-enabled tools. The report identifies cybercrime enablers, online fraud, cyber-attacks, and online child sexual exploitation as major areas of concern, with AI making scams, deception, and abuse more scalable and convincing.

Norway has announced plans to introduce a ban on social media use for children under 16, placing responsibility for age verification on technology companies.

Greece is moving to tighten restrictions on minors’ use of social media, with legislation expected later this year that would introduce a ban for children under 15. The measure is set to take effect on 1 January 2027 and is intended to be a framework that changes how platforms operate. Platforms would be required to implement robust age verification mechanisms, including the re-verification of existing accounts, with oversight provided by national regulators such as the Hellenic Telecommunications and Post Commission (EETT).

French President Emmanuel Macron is convening EU leaders, including Spanish Prime Minister Pedro Sanchez and representatives of Italy, the Netherlands and Ireland, to align national approaches to restricting minors’ access to social media and to press for faster EU-level action.

The UK’s Children’s Wellbeing and Schools Bill is set to expand ministers’ powers to shape how online services protect children, including by restricting access to risky platforms, features, or functions and by targeting design elements such as contact settings, live communication, location visibility, and time spent online. The draft would also bring Ofcom into a stronger advisory role, introduce a six-month timeline for regulations or a progress update, and give ministers new authority over children’s data consent, age assurance, and enforcement. The regulatory package remains unsettled for now, with Parliament still negotiating key provisions and no final law yet in place.

The European Commission has developed a standardised age-verification app intended to work across member states. The app allows users to confirm they meet age requirements to access social media platforms by providing their passport or ID number. It is designed to integrate into national digital wallets or operate as a standalone app, with a coordinated EU framework to ensure interoperability and avoid fragmented national systems. The app is open source and available for both public and private implementation, but is subject to common technical and privacy requirements. The Commission plans to establish an EU-level coordination mechanism to oversee rollout, accreditation, and cross-border usability. The rollout has faced scrutiny by security researchers. Reported weaknesses include locally stored authentication data that can be reset or modified, allowing users to bypass PIN protections, disable biometric checks, and reset rate-limiting mechanisms by editing configuration files. This effectively enables the reuse of verified identity data under altered access controls. The criticism has triggered broader concerns among developers about the app’s architecture, including why secure hardware features were not used, and whether elements like expiring age credentials are logically necessary.

The European Commission has also recently taken preliminary action against Meta, finding that Facebook and Instagram have not effectively prevented users under 13 from accessing their services, largely because age checks can be bypassed with false birthdates and weak verification systems.

Australia’s child-safety push is widening from social media to gaming, as regulators intensify scrutiny of how platforms protect minors from harm. On 21 April, the eSafety Commissioner issued legally enforceable transparency notices to Roblox, Minecraft, Fortnite and Steam, demanding details on how they handle risks, including child sexual exploitation, cyberbullying, hate and extremist material on services widely used by children.

The UK Information Commissioner’s Office has launched a campaign to help parents and carers speak with primary school-aged children about online privacy, after research found that many children are sharing personal details online, while families often feel unsure how to respond. The ICO says 24% of children have shared their real name or address online, 22% have disclosed information such as health details to AI tools, and 21% of parents have never discussed online privacy with them.

Economic

The European Commission has issued a supplementary charge sheet to Meta (called Supplementary Statement of Objections), outlining concerns over potential restrictions on third-party AI assistants’ access to WhatsApp. Previously, Meta decided to reinstate access to WhatsApp for third-party AI assistants for a fee. However, the Commission has preliminarily found that these measures remain anticompetitive and has now issued interim measures to prevent these policy changes from causing serious harm on the market. The interim measures would stay in effect until the Commission concludes its investigation and issues a final decision on Meta’s conduct.

UNCTAD reports that global trade grew by $2.5 trillion in 2025 to reach $35 trillion, reflecting continued expansion in goods and services but also a more fragile and uneven economic landscape. Rising geopolitical tensions, disrupted shipping routes, conflicts in the Middle East, and instability in key maritime corridors are driving up energy, transport, and import costs, placing heavier pressure on developing economies with limited fiscal space. Services growth has slowed, while much of the recent trade increase stems from higher prices rather than stronger volumes. East Asia and Africa remain important drivers through South–South trade and shifting supply chains, yet fragmentation, US–China decoupling, inflation, debt, and protectionism are expected to weigh on 2026 prospects.

The International Labour Organisation warns that social protection systems are failing to keep pace with fast-changing labour markets shaped by climate change, technological disruption, demographic shifts, and evolving forms of work. Its new report highlights major gaps in coverage, adequacy, and financing, leaving many workers exposed during unemployment, illness, retirement, or job transitions.

Russia is moving to criminalise large-scale unauthorised cryptocurrency activity, after a government legislative commission approved amendments that create prison sentences for organising the circulation of digital currency without central bank authorisation. The proposed Article 171.7 of the Criminal Code would punish cases involving significant harm, major illicit income, or damage to individuals, organisations, or the state with a sentence of 4 to 7 years in prison. Expected to take effect on 1 July 2027, the measure marks a sharper enforcement turn in Russia’s digital asset sector.

The European Commission has updated its technology transfer competition rules to reflect better data-driven innovation, digital markets, and modern licensing practices across the EU. The revised framework clarifies how companies can license patents, software, know-how, and data-related technologies while staying within competition law, aiming to protect collaboration and legal certainty without allowing agreements that restrict market access or innovation. Greater attention is given to digital ecosystems, standard-essential technologies, and licensing arrangements that may shape control over data, interoperability, and downstream competition.

Canada has announced C$23.8 million for the Digital Skills for Youth programme, aiming to help young people gain practical experience as AI, cybersecurity, big data, automation, and broader digital transformation reshape the labour market. Led by Industry Minister Mélanie Joly, the two-year initiative will fund training and work placements for post-secondary graduates by linking them with employers across emerging technology sectors. Eligible recipients include businesses, non-profits, public institutions, Indigenous organisations, and provincial or territorial bodies, with flexible access for participants in Yukon, the Northwest Territories, and Nunavut.

Human rights

Brazil has inaugurated its first Center for Access, Research and Innovation in Assistive Technology (Capta) at the Benjamin Constant Institute in Rio de Janeiro. Run by the Ministry of Science, Technology and Innovation (MCTI) under the National Plan for the Rights of People with Disabilities, the centre aims to foster the development, experimentation, and dissemination of assistive technologies that enhance autonomy, inclusion, and quality of life for people with disabilities. The launch marks the first of several planned centres nationwide to expand access to these technologies.

UNESCO warns that students with disabilities continue to face deep barriers in education, including inaccessible infrastructure, limited assistive technologies, insufficient teacher training, stigma, and weak data systems that leave many learners invisible in policy planning. Its findings show that exclusion often begins early and is reinforced by poverty, gender inequality, displacement, and other overlapping disadvantages, limiting access to quality learning and future opportunities. UNESCO urges governments to move beyond narrow inclusion measures by investing in accessible schools, inclusive curricula, trained educators, reliable data, and meaningful participation by persons with disabilities.

The Philippines and Bermuda have signed a memorandum of understanding to strengthen cross-border cooperation on personal data protection, linking the Philippines’ National Privacy Commission with Bermuda’s Office of the Privacy Commissioner. The agreement enables information sharing, mutual assistance in investigations, and closer coordination on data breach cases that cross jurisdictions. Beyond enforcement, the partnership supports compatible data protection mechanisms, certification frameworks, trusted data flows, training, and knowledge exchange on emerging privacy challenges.

Legal

A unanimous US Supreme Court ruling has narrowed the circumstances under which an internet service provider (ISP) can be held liable for users’ copyright infringement. Writing for the Court, Justice Clarence Thomas said an ISP is liable only if its service was designed for unlawful activity or if it actively induced infringement.

French authorities have summoned Elon Musk and former X chief Linda Yaccarino to give voluntary interviews in relation to a criminal investigation into whether X enabled the spread of child sexual abuse material, AI-generated deepfakes, Holocaust denial content, and other harmful or unlawful material. However, Musk appears to have refused by not showing up. The confrontation widened when reports emerged that the US Justice Department had declined to assist the French inquiry, arguing that the case risked crossing into the regulation of protected speech and that it would unfairly target a US company. French authorities, however, have framed the matter as a legitimate enforcement action under national law.

In the federal multidistrict litigation (MDL) pending in the Northern District of California involving Meta, Google (YouTube), ByteDance (TikTok), and Snap Inc. the court denied motions to dismiss filed by several school districts. That moves the case out of the pleading stage and into bellwether proceedings, where selected cases will test core liability and damages theories. The plaintiffs’ main argument is product design-based. They claim the platforms were engineered to maximise engagement among minors despite internal awareness of mental health risks. They link this to reported increases in anxiety, depression, and behavioural disruption in school environments. The causal chain is disputed, but that is the core theory being advanced. The MDL is large, with over 2,300 related actions across six states, making it one of the more significant litigations in this area. The upcoming June bellwether trial is expected to be the first real test of these claims and will likely influence both settlement pressure and the broader direction of the MDL.

Raine v. OpenAI is proceeding as a standalone case in California, not part of any MDL. The complaint alleges that Adam Raine’s use of ChatGPT shifted from academic purposes to emotional reliance, with escalating mental health disclosures allegedly met by responses that reinforced dependence rather than directing him outward. The plaintiffs argue this was a foreseeable result of engagement-oriented design. They bring claims including wrongful death and seek injunctive relief for stronger safeguards. While no trial date has been formally set, the case remains in its early procedural stage in California and may proceed toward trial in late 2026 or 2027, depending on pretrial developments.

Sociocultural

The European Commission has launched a Mediterranean digital transformation programme for North African and Middle Eastern countries, marking the first digital initiative under the Pact for the Mediterranean. The programme aims to support inclusive and sustainable growth by improving access to digital services, aligning telecommunications regulation with the EU standards, and strengthening national regulatory authorities. Cybersecurity is a core priority, with support for stronger national frameworks, institutional capacity, and coordinated responses to digital threats.

The European Commission’s first monitoring results under the revised Code of Conduct on Countering Illegal Hate Speech Online+ show that major platforms are making progress in reviewing reported illegal hate speech within 24 hours, while gaps remain in accuracy, consistency, and reporting practices. Based on independent monitoring and company data, the assessment found that many notifications were handled within agreed timelines, but a notable share of cases were disputed or wrongly classified. Linked to the Digital Services Act’s co-regulatory model, the exercise acts as a practical test of platform accountability, transparency, and compliance with the EU and national law.

The UK government is planning measures that could make senior technology executives face criminal charges, including prison sentences, if their companies fail to remove non-consensual intimate images when required by regulators. The move builds on existing obligations that already require platforms to take down such material within strict timeframes or face significant penalties, including fines of up to 10% of global turnover or even service blocking. The latest step goes further: instead of relying solely on corporate sanctions, it introduces personal criminal accountability at the executive level. This type of liability is likely to accelerate compliance in ways that financial penalties alone have not, and may serve as an example to other jurisdictions. The policy is part of a broader tightening of the UK’s online safety framework, driven by persistent concerns over revenge porn and the rapid proliferation of AI-generated intimate imagery.

AI governance in April

National plans and initiatives

India. India has set up a Technology and Policy Expert Committee under the Ministry of Electronics and Information Technology to help shape the country’s AI governance framework and advise the new AI Governance and Economic Group. Bringing together government, academia, industry, and policy expertise, the body is meant to translate fast-moving technical and regulatory issues into practical guidance, bringing a more structured and adaptive approach to AI governance aligned with India’s economic and social priorities.

South Africa. South Africa has withdrawn its draft national AI policy after it was discovered that the document contained fake, AI-generated citations, undermining the credibility of the proposed framework. The government said the lapse occurred due to a failure to verify references and stressed that stronger human oversight is required in policy processes involving AI tools. The withdrawal delays plans to establish new AI governance institutions and incentives, and the policy will now be redrafted.

Sovereignty

The UK. The government is planning to back British strengths in the parts of the AI stack where the UK can build real leverage, Liz Kendall, the UK’s Secretary of State for Science, Innovation and Technology, stated. Kendall rejected technological isolationism, instead championing AI sovereignty for Britain: reducing over-dependencies, backing domestic firms with a £500 million Sovereign AI fund, and launching a new AI Hardware Plan in June 2026 to capture chip market share. Kendall also advocated collaboration with other middle powers, including on setting the standards for how AI is deployed.

The government has also launched a £500 million Sovereign AI Fund to accelerate domestic AI startups and strengthen national technological autonomy. The initiative combines direct equity investment with access to national compute infrastructure, fast-tracked visas for global talent, and procurement pathways into public services. It targets early-stage to growth companies in areas such as AI infrastructure, life sciences and advanced computing, with the explicit goal of ensuring that high-potential firms scale and remain anchored in the UK rather than relocating abroad.

Papua New Guinea. The government has issued new guidance on AI and data sovereignty, setting out principles for ensuring that national data assets remain under domestic control. The framework emphasises governance over data storage, processing and cross-border transfers, particularly where public-sector or sensitive datasets are involved.

Russia. Russia is advancing a draft AI regulatory framework that would formalise oversight of AI development and deployment, aligning with broader efforts to strengthen digital sovereignty and state control over emerging technologies. The proposals focus on risk management, national standards and reducing dependence on foreign AI systems, while supporting domestic innovation. The move fits into Moscow’s broader strategy to tighten control over digital infrastructure and cross-border data flows.

Partnerships

South Korea–France. South Korea and France are deepening cooperation through a new strategic AI and technology partnership, aimed at strengthening joint research, industrial collaboration and standard-setting across emerging technologies. The initiative reflects a broader effort to align capabilities in semiconductors, data infrastructure and advanced computing, while positioning both countries more competitively in the global AI landscape.

The EU-Morocco. The European Commission and Morocco have launched a digital dialogue to deepen strategic cooperation on emerging technologies, digital transformation, and innovation-led development. Focused on AI, digital infrastructure, start-up support, research collaboration, and stronger AI ecosystems, the initiative aims to turn digital technologies into drivers of economic and social progress. Greater interoperability of digital public services and expanded knowledge exchange are also central to the partnership, reflecting a shared interest in more connected, efficient, and inclusive digital governance.

Legal

The USA. A federal appeals court in Washington, D.C. has declined to block the Pentagon’s national-security blacklisting of Anthropic, allowing the designation to remain in force while litigation continues. The ruling contrasts with a separate decision by a California judge who had earlier blocked part of the government’s action, highlighting a growing judicial split over the unprecedented move.

Paraguay. Paraguay has adopted new rules for the use of AI in its courts, with UNESCO support, marking a notable step in judicial AI governance. The framework, approved by the Supreme Court of Justice, limits AI to a supporting role in data processing, information management, and assisted decision-making, while requiring human oversight, transparency, accountability, and disclosure when AI tools influence judicial processes. The rules align Paraguay’s approach with UNESCO’s guidance on AI in courts and underscore a wider trend toward rights-based, trust-focused AI deployment in public institutions.

Belgium. Belgium’s data protection authority has released a new information brochure titled ‘The Impact of Artificial Intelligence (AI) on Privacy’, providing guidance on risks such as bias, privacy violations and misuse of generative AI systems. The document is intended to raise awareness among organisations and the public, and to support compliance with EU data protection and AI governance frameworks.

Safety and security

The EU. EU member states and European Parliament lawmakers have failed to reach an agreement on revisions to the EU Artificial Intelligence Act, after 12 hours of negotiations over proposed changes under the Commission’s Digital Omnibus package. Disagreements centred on whether sectors already covered by existing product and safety regulations should be exempt from certain parts of the AI framework. Lawmakers warned that the latest deadlock risks creating legal uncertainty for companies already preparing for compliance, while privacy and civil society groups cautioned that proposed relaxations could weaken core safeguards. Talks will, however, resume in May.

Kazakhstan. Kazakhstan has introduced mandatory audits for high-risk AI systems, requiring developers to obtain a positive audit assessment before their systems can be listed as ‘trusted’ by sectoral authorities. The government will publish and regularly update official lists of approved systems, based on applications that include documentation on ownership, functionality and use conditions, reviewed within strict timelines. The move aims to build trust and standardise best practices in AI deployment, signalling a more structured and compliance-driven approach to high-risk AI governance.

New Zealand, the UK, Singapore. New Zealand’s National Cyber Security Centre, the UK National Cyber Security Centre, and Singapore’s Cyber Security Agency have issued coordinated warnings that frontier AI is reshaping the cyber threat landscape by lowering barriers to sophisticated attacks, accelerating vulnerability discovery, and compressing the window between disclosure and exploitation. All three stress the dual-use nature of AI, urging organisations to reassess outdated risk models and prioritise rapid patching, continuous monitoring, stronger identity and access controls, and reduced attack surfaces to counter increasingly automated and faster-moving cyber threats across both public and private sectors.

The USA. US cybersecurity officials are considering reducing the patching deadline for actively exploited flaws to just three days, citing the accelerating speed at which AI systems can identify and weaponise vulnerabilities. The proposed shift would initially apply to federal civilian agencies but could redefine baseline incident response expectations across government and critical infrastructure, with agencies arguing that traditional patch cycles are no longer compatible with current exploit timelines, while industry warns that such compressed deadlines may exceed the capacity of complex and legacy IT environments.

Meanwhile, Washington is quietly reversing course on its standoff with Anthropic. The White House is drafting executive guidance that would allow federal agencies to work with Anthropic again, despite the company previously being labelled a supply-chain risk by the Pentagon. The shift reflects internal fractures: while parts of the defence establishment remain wary, others see excluding frontier models like Mythos as strategically costly.

Mythos. Anthropic has launched an investigation after a small group of users gained unauthorised access to its powerful Mythos AI model via a third-party contractor environment. The access reportedly occurred just as the company began rolling out a limited preview of the model to selected organisations under Project Glasswing. The unauthorised users are believed to have operated through a private Discord group, using a mix of tactics, including contractor access and open-source intelligence tools, to gain access to the system. Mythos was intentionally restricted due to its ability to accelerate cyberattacks and was provided to a limited number of partners, yet it appears to have leaked almost immediately through the partner ecosystem rather than through a direct breach. The window during which Mythos’ capabilities remain contained may prove far shorter than anticipated.

Content governance

China. The Cyberspace Administration of China has warned several ByteDance-owned platforms, including CapCut, Catbox and the Dreamina AI system, over failures to properly label AI-generated and synthetic content. The Cyberspace Administration of China said inspections found violations of cybersecurity and generative AI regulations, prompting enforcement measures such as mandatory rectification, warnings and disciplinary action against responsible personnel.

Development

Ghana. The Ghanaian Ministry of Communication, Digital Technology and Innovations has launched a public-sector AI capacity development programme in collaboration with the Government of Japan and the UN Development Programme. The programme is designed to equip public officials with knowledge of AI and its applications in governance. It focuses on improving decision-making and service delivery, drawing on experience from the UN and Japan.

UNESCO-Latin America & Caribbean. UNESCO has launched a regional AI in Education Observatory for Latin America and the Caribbean, designed to support evidence-based policymaking and track the impact of AI on education systems. The initiative aims to build capacity, share best practices and guide responsible integration of AI tools in schools and learning environments.

UNESCO–Oxford. UNESCO and the University of Oxford have launched a global AI course for courts. The programme trains judges and legal professionals to assess algorithmic tools, identify bias, and ensure compliance with human rights standards in increasingly digitalised judicial processes. It introduces practical frameworks for evaluating AI outputs in legal contexts, with a strong focus on maintaining judicial independence, transparency and accountability as AI becomes embedded in evidence handling and decision-support systems.

Commonwealth. The Commonwealth Secretariat has launched a capacity-building programme on the use of AI in election management, training electoral officials from member states on how AI tools can support voter education, administrative efficiency and data analysis while safeguarding electoral integrity. The initiative focuses on practical applications of AI in electoral processes, including risks such as misinformation, bias and automation of sensitive decision-support functions. It emphasises that AI should remain assistive rather than substitutive in democratic processes, with human oversight positioned as central to maintaining trust, legitimacy and accountability in elections.

Australia. Under its national AI workforce strategy, Australia is expanding targeted upskilling programmes for learners and workers to address structural skill gaps created by AI-driven labour market shifts. The approach prioritises integration of AI literacy into education and vocational pathways, alongside employer-linked training to support adaptation in high-exposure sectors. It frames AI as a general-purpose technology requiring continuous reskilling rather than one-off training, with policy attention on inclusion, transition support and alignment between education systems and emerging digital economy demands.

Pakistan. Pakistan has approved the establishment of an AI Education Authority alongside plans for virtual schools. The reforms aim to scale AI-driven learning systems, support personalised education delivery and standardise digital curricula across regions. The initiative is framed within broader efforts to modernise the education sector, strengthen digital access, and build national capacity for AI adoption in public education, while addressing disparities in learning outcomes through technology-enabled delivery models.

When AI writes the rules: How to avoid fake laws governing real life

Last month, South Africa unveiled its first draft national AI policy, aiming to position the country as a continental leader in innovation. The plan included ambitious new institutions: a National AI Commission, an Ethics Board, and tax breaks for private sector collaboration.

But just days later, the celebration turned sour.

According to Reuters, South Africa’s government was forced to withdraw the draft after reviewers discovered a fatal flaw: the policy was riddled with fake resources and citations that didn’t exist. The research supporting the country’s AI strategy had likely been generated by an AI.

This isn’t a minor typo. AI hallucinated policies and supporting resources. It is not surprising, as LLMs are advanced guessing machines, not providers of verified facts. Even when they are fake, texts can look perfectly correct and legitimate.

South Africa’s Minister of Communications and Digital Technologies, Solly Malatsi, acknowledged the failure with refreshing honesty:

‘The most plausible explanation is that AI-generated citations were included without proper verification. This should not have happened.’

He noted that this lapse ‘has compromised the integrity and credibility of the draft policy.’

Why does it matter? We are not highlighting South Africa to single it out or cause embarrassment. We are shining a spotlight on the problem with AI-generated laws. South Africa’s incident is not an exception. As policymakers rush to keep up with technology, we are seeing more examples of AI-drafted regulations being submitted for review. For instance, in the USA, a federal judge in California sanctioned two law firms for submitting a legal brief containing a fake citation generated by AI.

The problem isn’t that AI is used. The danger lies in how it is being used.

Legal documents and policies require precision, grounding, and contextualisation. Generic AI models often fail at all three:

Lack of precision: AI frequently provides vague, generic answers to specific legal questions. Laws need pointed, solid definitions; AI prefers probabilistic guesswork.
No grounding: Most AI models cannot provide a verifiable link to the exact sentence of a law or regulation. Often, they mix up jurisdictions across countries and jurisdictions.
Zero context: AI frequently lacks the specific political, social, or historical context of a policy and regulations. Temporal context is also missing, which shows how legal issues evolved over the course of drafting and negotiations.

How to fix a problem (without banning AI). The solution lies in a two-pronged approach: developing institutional AI and increasing AI literacy.

If South Africa had had institutional AI anchored into local knowledge and context, such a hallucination could have been avoided. Moreover, AI would be a genuine and useful tool reflecting the topical and temporal context of policy development and law drafting.

But more importantly, we need to build AI competencies among policymakers. This requires a shift in pedagogy. We cannot teach policymakers to simply use AI; they must understand how it works.

As Minister Malatsi stated:

‘This unacceptable lapse proves why vigilant human oversight over the use of artificial intelligence is critical. It’s a lesson we take with humility.’

If we fail to build precise, grounded AI tools and train policymakers to use them properly, we won’t just have fake citations in a draft. We will have fake laws governing real people.

Read the original ‘When AI writes the rules: How to avoid fake laws governing real life’ blog post by Dr Jovan Kurbalija.

Designing our future

April brought another wave of high-profile predictions. While some might be tempted to dismiss them as mere hype, there’s a strong reason not to. These ideas come from people who are not only building the platforms or technologies we rely so much on, but are also spending capital to transform into reality their visions of what the future should look like. When ‘tech leaders’ float their ideas, they begin to steer real-world resources and regulatory conversations. And there we come to the quiet danger: designing the future without meaningful public input.

The co-founder of Palantir, Alex Karp, and Palantir’s Head of Corporate Affairs, Nicholas W. Zamiska, published a set of 22 propositions drawn from their upcoming book, Technological Republic. It did not arrive quietly. Critics called it ‘technofascism’ and ‘what evil would tweet.’

Their vision is organised around duty, hard power, and scepticism toward modern democratic culture. They argue that Silicon Valley owes a moral debt to the country that made its rise possible, and that the engineering elite has an affirmative obligation to participate in national defence. They question the all‑volunteer force, suggesting that national service should be a universal duty so that the next war involves shared risk. Soft power and soaring rhetoric, they writes, have been exposed as insufficient. Free societies need hard power, and in this century, hard power will be built on software.

When it comes to AI weapons, Karp and Zamiska are blunt: they will be built regardless of Western debates. The only question is by whom and for what purpose. The authors also defend Elon Musk against what he sees as cultural snickering, arguing that we should applaud those who attempt to build where the market has failed to act. At the same time, they reject what they calls vacant pluralism, insisting that not all cultures are equally productive and that elite intolerance of religious belief is a sign of intellectual closure.

What Kapr and Zamiska do not offer is much economic policy. Their technological republic is organised around security and technological power, not redistribution. The state exists to be defended. The individual exists to serve.

Book, Comics, Publication, People, Person, Adult, Male, Man, Face, Head, Jury, Crowd, José Luis Álvarez Álvarez

Around the same time, OpenAI released its own policy document, Industrial Policy for the Intelligence Age. It is longer, somewhat softer, full of phrases like ‘public wealth fund’ and ‘right to AI.’ It asks for a democratic conversation about AI industrial policy, regulation, ethics and economy.

OpenAI’s document starts from a different problem. Superintelligence—AI systems capable of outperforming the smartest humans, even when those humans are assisted by AI—is coming. Market forces alone cannot manage the transition, OpenAI argues. Drawing parallels to the Progressive Era and the New Deal, the company proposes ambitious public‑private collaboration.

On the economic side, this includes giving workers a formal voice in how AI is deployed in workplaces, microgrants to help workers become AI‑first entrepreneurs, a right to AI as foundational access comparable to literacy or electricity, shifting taxation from payroll to capital gains and automated labour, creating a Public Wealth Fund to give citizens a direct stake in AI‑driven growth, and converting efficiency gains into shorter workweeks or better benefits.

On the resilience side, OpenAI proposes safety systems for cyber and biological risks, an AI trust stack for verification, auditing regimes for frontier models, model‑containment playbooks for dangerous AI, and guardrails for government use. The company acknowledges it does not have all the answers and invites feedback.

Similarities and differences. Where Karp and Zamiska talk about duty and war, OpenAI talks about transitions and safety nets. Yet both reject the current political order as inadequate. Both see technology as the primary vector of power. And both propose new forms of obligation—national service in one case, a right to AI and portable benefits in the other.

Taken together, these two documents are not opposing manifestos. They are different dialects of the same emerging language: tech leaders no longer see themselves as toolmakers. They see themselves as institutional designers. And a courtroom battle between Elon Musk and Sam Altman is about to decide how enforceable their original promises really are.

Promises, promises. Elon Musk is suing Sam Altman over whether OpenAI was fraudulently diverted from its original nonprofit mission. Musk argues that he was misled and that OpenAI’s leadership abandoned its promise to serve humanity, pivoting instead toward commercialisation through partnerships and products like ChatGPT. He seeks to remove Altman and President Greg Brockman, force structural changes to OpenAI’s governance, and potentially award up to $150 billion in damages to its nonprofit arm. OpenAI rejects this narrative, framing the case as a competitive dispute—Musk raised objections, they say, only after OpenAI’s success and the emergence of his own AI venture, xAI, which has filed for an IPO. OpenAI itself is rumoured to be considering an IPO in late 2026 or 2027. The court will have to weigh early emails, funding discussions, and conflicting interpretations of what “open” and “nonprofit” were supposed to mean.

If the court rules that shifting toward profit violated founding principles, many similar hybrid organisations may need to restructure. If the current model is upheld, it will solidify the reality that market logic and commercial interest drive AI development. Because advanced AI is expensive to build and operate, companies need pricing tiers to cover costs and make a profit. And because the underlying models and infrastructure are valuable competitive assets, firms have incentives to lock users in and limit disclosure to maintain their advantage. That means that users could be facing more tiered access, stronger platform lock‑in, and less visibility into how systems operate.

So what could societies do? Karp and Zamiska, and OpenAI share a premise that is rarely stated outright: that the existing legal and political order is too slow or too confused to manage the technologies now emerging.

If we assume they are even only partially right, the solution cannot be handing design authority to the same firms that profit from those technologies. Three measured steps are worth considering.

First, separate policy design from corporate strategy. Any company that holds major public contracts in areas such as defence, health, or border control should not be the source of the policies used to regulate that company’s activities.

Second, codify accountability. If AI developers claim public-interest missions, those claims need legal and regulatory grounding, not just branding. The Musk-OpenAI case may accelerate this, but policymakers cannot outsource the task to courts.

Third, broaden participation. OpenAI’s call for public input points in the right direction, but mechanisms matter. Without meaningful inclusion—across labour, civil society, and smaller economies participation risks becoming procedural rather than substantive.

We are not about to wake up in a technological republic overnight. But it is already clear that tech oligarchs are no longer just building products; they are articulating political and social orders. Modern societies will have to get right what type of legal and policy order is needed and how to deal with the growing power of tech companies and their leaders.

Claude Mythos Preview sets new benchmark for AI capability and raises governance questions

On 7 April 2026, Anthropic announced Claude Mythos Preview, its most capable AI model to date, alongside the explicit decision not to make it publicly available. Claude Mythos Preview is a general-purpose, unreleased frontier model that, in Anthropic’s own words, reveals a stark fact: AI models have reached a level of coding capability where they can surpass all but the most skilled humans in finding and exploiting software vulnerabilities.

Anthropic’s published benchmarks show Mythos Preview scored 93.9% on the SWE-bench Verified test, 97.6% on the USAMO 2026 mathematics evaluation, and and significantly outperformed all previously released models in cybersecurity-specific assessments. The SWE-bench Verified score is roughly double the 2024 state of the art and was achieved in an agentic context, where the model autonomously resolved real software engineering issues from production codebases.

On the USAMO 2026 evaluation, Mythos Preview scored 55 percentage points higher than Opus 4.6, which scored 42.3%. On GPQA Diamond, a graduate-level scientific reasoning benchmark, Mythos Preview scored 94.6%. On Terminal-Bench 2.0, which evaluates system administration and command-line proficiency, it scored 82.0%, a 16.6-point lead over Opus 4.6. On the cybersecurity benchmark Cybench, the model scored 100% on the first attempt, making it no longer useful as a discriminating evaluation.

Cybersecurity capabilities

The decision not to release Mythos Preview publicly is linked to concerns about its advanced capabilities, particularly in high-risk domains such as cybersecurity, as well as broader considerations related to safety and potential misuse.

Notably, these capabilities are not the result of targeted training. Anthropic did not explicitly train Mythos Preview to have these capabilities. They emerged as a downstream consequence of general improvements in code, reasoning, and autonomy. The same improvements that make the model substantially more effective at patching vulnerabilities also make it substantially more effective at exploiting them.

During internal testing, Mythos Preview identified thousands of zero-day vulnerabilities across every major operating system and every major web browser, as well as other critical software, many of them high severity and previously undetected for years. Anthropic engineers with no formal security training could ask Mythos to find remote code execution vulnerabilities overnight and have a complete, working exploit the following morning. This accessibility dimension poses a distinct governance concern. Traditionally, sophisticated cyberattacks have required highly skilled teams, extensive planning, and deep technical expertise. Models with these capabilities may lower those barriers substantially, including smaller state actors and non-state actors.

Anthropic has disclosed only a fraction of what it says it has found during internal testing. Over 99% of the vulnerabilities discovered by Mythos remained unpatched at the time of the 7 April announcement.

Project Glasswing

Anthropic launched Project Glasswing as a structured access mechanism to use Claude Mythos Preview for defensive cybersecurity purposes. The initiative brings together Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks as launch partners, with access also extended to over 40 additional organisations that build or maintain critical software infrastructure.

Project Glasswing partners will receive access to Claude Mythos Preview to find and fix vulnerabilities in their foundational systems, with work expected to focus on local vulnerability detection, black box testing of binaries, securing endpoints, and penetration testing. Anthropic is committing up to $100M in usage credits for Mythos Preview across these efforts. Following the initial research preview period, access to the model will be available to participants at $25 per million input tokens and $125 per million output tokens across the Claude API, Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry.

Art, Graphics, Advertisement, Poster, Text

Anthropic has also donated $2.5M to Alpha-Omega and OpenSSF through the Linux Foundation, and $1.5M to the Apache Software Foundation to enable open-source software maintainers to respond to the changing cybersecurity landscape.

Within 90 days, Anthropic has committed to reporting publicly on what it has learned, as well as the vulnerabilities fixed and improvements made that can be disclosed. The company also intends to collaborate with leading security organisations to produce practical recommendations covering vulnerability disclosure processes, software update processes, open-source and supply-chain security, and patching automation, among other areas.

Anthropic has stated that Project Glasswing is a starting point, and that in the medium term, an independent, third-party body bringing together private and public sector organisations might be the ideal home for continued work on large-scale cybersecurity projects.

Project Glasswing raises a governance question for the industry, as cyber-capable AI systems may become useful security tools and a source of misuse risk at the same time. Project Glasswing’s structure also reveals tensions, as it concentrates several roles including discovery, disclosure coordination, and capability gatekeeping in a single organisation. Entities such as Anthropic and major cloud providers control critical components of the Glasswing ecosystem, raising questions about power and governance that, for financial institutions in particular, translate into systemic risk.

We also wrote about the Glasswing project and its implications in our Weekly newsletter in early April.

Geopolitical dimensions

Claude Mythos has sharpened attention on the competitive and geopolitical dimensions of frontier AI development. Project Glasswing’s launch partners exclude Anthropic’s rival OpenAI, which is reported to be approximately six months behind Anthropic in developing a model with comparable offensive cyber capabilities.

Senior policy voices have positioned Mythos within the broader competition between Western AI companies and China’s rapidly evolving AI ecosystem, with implications for national security, enterprise adoption, and technological leadership. A security researcher assessed a concurrent source code leak from Anthropic as a geopolitical accelerant, noting that such exposures compress the timeline for adversaries to replicate technological advantages currently held by Western laboratories.

Many defence organisations still rely on legacy software and infrastructure not designed with AI-driven threats in mind. Models capable of autonomously identifying hidden flaws in older code may expose weaknesses in critical defence networks around the world. The difficulty of containment at the geopolitical level is reflected in usage patterns. Access restriction at the laboratory level does not translate reliably into containment across jurisdictions when the same underlying models are accessible via cloud infrastructure spanning multiple countries and regulatory environments.

The limits of voluntary AI governance

The Claude Mythos case has clarified, with considerable precision, what voluntary AI governance can and cannot achieve. A responsible laboratory can make a unilateral decision not to release a dangerous system. It can support coordinated vulnerability disclosure, engage governments proactively, and produce detailed public documentation of a model’s capabilities and risks. All of these have occurred with Mythos, and represent meaningful progress relative to the governance environment of a few years ago.

What voluntary frameworks cannot do is bind competitors who operate under different assumptions. Anthropic’s RSP version 3.0 acknowledges this directly by removing the commitment to withhold unsafe models if another laboratory releases a comparable model first. The competitive structure of the AI industry means that restraint by one actor does not prevent the underlying capability from eventually proliferating. Voluntary governance frameworks work best when they generate shared norms across an industry. When the industry is structured around intense competition among a small number of organisations, voluntary restraint by a single actor does not resolve the broader question of access.

Analysts note that what Mythos does today in a restricted environment, publicly available models are likely to replicate within one to two model generations. The next phase of the EU AI Act takes effect in August 2026, introducing automated audit trails, cybersecurity requirements for AI systems classified as high risk, incident reporting obligations, and penalties of up to 3% of global revenue. The EU framework represents a shift toward binding governance, but its scope relative to the pace and international distribution of frontier AI development remains to be demonstrated.

The way forward

Anthropic acknowledges that capabilities like those demonstrated by Mythos will proliferate beyond actors committed to deploying them safely, with potential fallout for economies, public safety, and national security. The company’s response, taken in aggregate, reflects a serious attempt to manage that risk within the constraints of voluntary frameworks and private decision-making. The Responsible Scaling Policy, Project Glasswing, proactive government briefings, and the detailed system card are each substantive contributions. They are also all products of a single private entity’s judgement, operating without binding external accountability.

The Mythos case does not so much call for a different assessment of Anthropic’s conduct as it does a clear-eyed view of what voluntary governance can realistically sustain at the frontier of AI development. Governments on both sides of the Atlantic were briefed informally about a model whose capabilities are consequential for critical infrastructure and national security. No binding notification requirement existed. No independent technical authority had prior access. No international coordination mechanism was in place.

No single organisation can solve these challenges alone. Frontier AI developers, software companies, security researchers, open-source maintainers, and governments all have essential roles to play. The Mythos case has made that observation not merely a statement of aspiration but a policy problem that requires concrete institutional responses. Whether those responses will take shape before the next capability threshold is reached is the question now facing policymakers.

This text is an adaptation of Reyhan Damalan’s text ‘Claude Mythos Preview sets new benchmark for AI capability and raises governance questions‘.

Last month in Geneva

29th session of the CSTD

The 29th session of the Commission on Science and Technology for Development (CSTD) took place from 20 to 24 April 2026 at the Palais des Nations in Geneva, Switzerland.

For its 29th session, the programme addressed the priority theme of ‘Science, Technology and Innovation in the Age of Artificial Intelligence’ and heard presentations on report on technical cooperation activities in science, technology and innovation.

CSTD members also reviewed progress in implementing and following up to the outcomes of the World Summit on the Information Society at the regional and international levels.

The session also included a briefing on the plans of the International Scientific Panel on AI, a report on Global Digital Compact (GDC) implementation milestones, as well as an opportunity to engage with the Draft Joint Implementation Road Map for WSIS-GDC Coherence.

Ultimately, CSTD members adopted two resolutions on WSIS and Science, Technology & Innovation for Development.

Image credit: UNCTAD Innovation X post

Shaping Switzerland’s AI Summit Strategy

A report intended to inform strategic planning for the AI Summit Geneva 2027, synthesising inputs from a multistakeholder roundtable and 50+ written submissions to shape Switzerland’s strategy for hosting the AI Summit, has been made public.

The core finding of ‘Shaping Switzerland’s AI Summit Strategy’ is that Switzerland’s comparative advantage lies not in technological scale, but in trusted convening, pragmatic governance, and institutional credibility. Its neutrality, strong institutions, research base (e.g. ETH/EPFL), and Geneva’s multilateral ecosystem position it as a facilitator of practical, cross-sector cooperation. However, gaps remain in investment and in scaling innovations to market.

Two priority issue clusters dominate. First, trusted and sovereign AI infrastructure, including open models, interoperability, and reducing dependence on dominant providers—alongside a noted gap in Switzerland’s access to production-grade AI compute. Second, AI’s impact on human rights, security, and humanitarian law, particularly in relation to military use, surveillance, and preservation of human agency. Cross-cutting concerns include AI literacy, SME adoption, public-sector readiness, and equitable access for developing countries.

Strategically, Geneva 2027 should be framed as a platform for implementation, contributors highlighted, delivering a limited set of practical, internationally reusable tools backed by an inclusive preparatory process and follow-up mechanisms.

Geneva Cyber Week 2026

The UN Institute for Disarmament Research (UNIDIR) and the Swiss Federal Department of Foreign Affairs (FDFA) are co-hosting Geneva Cyber Week from 4 to 8 May 2026, bringing policymakers, diplomats, technical experts, industry leaders, academics, and civil society representatives to venues across Geneva and online for a week of discussions on cyber stability, resilience, governance, digitalisation, and the security implications of emerging technologies, including AI.
Returning after its inaugural edition, the event is being positioned as a response to a more fragile cyber and geopolitical environment. Held under the theme ‘Advancing Global Cooperation in Cyberspace’, Geneva Cyber Week 2026 comes at a moment of mounting cyber insecurity, intensifying geopolitical tension, and rapid technological change. The programme will feature nearly 90 events and reinforce Geneva’s role as a centre for cyber diplomacy, international cooperation, and digital governance.

Opportunity: Become a Knowledge Fellow

Diplo is pleased to launch a new call for applications for Digital Watch Knowledge Fellows (2026).

What is the Digital Watch Observatory?

Its content is organised around:

Topics, from cybercrime and freedom of expression, to data governance and critical infrastructure.
Technologies such as artificial intelligence, quantum computing, and semiconductors.
Processes including the UN Global Mechanism on ICT Security, the Internet Governance Forum, the Global Digital Compact process, and more.
Policy players such as countries, technical entities, business associations, UN entities, and other international and regional organisations.
Resources, including conventions, resolutions, laws and regulations, reports, and more.
Events, such as meetings, negotiations, conferences, and consultations.

This structure is complemented by:

Daily updates, regular analyses, and weekly and monthly newsletters that track and explain the most relevant developments across the digital governance landscape.

What is the role of a Knowledge Fellow?

Knowledge Fellows (KF) are central to the observatory’s ability to provide comprehensive, accurate, and up-to-date coverage of specific areas of digital governance.

Each KF is expected to cover one or more areas of expertise and help ensure that DW remains accurate, relevant, and complete, and impartial. This means:

Monitoring and analysing developments related to the assigned area(s) of expertise and ensuring these are reflected in daily updates and regular analyses.
Keeping assigned DW pages accurate, up-to-date, and substantively strong.
Tracking events relevant to their area(s) of expertise and helping ensure that important meetings, negotiations and discussions are reflected in DW.
Identifying key resources relevant to their area(s) of expertise such as UN resolutions and other intergovernmentally agreed documents, laws, regulations, reports, and policy papers.
Supporting stronger coverage of organisations, countries, and other key actors in digital governance.
Contributing, when relevant, to newsletters, policy and research papers, and other knowledge products.

Knowledge Fellows may also have opportunities to contribute to Diplo’s wider knowledge ecology, including courses, discussions, and thematic initiatives.

Who should apply?

At a time when the public space is abundant with AI-generated content, we are looking for more than just someone who can use AI to summarise news or rewrite online resources.

Specifically, we are looking for applicants who:

Have a strong expertise in digital governance, grounded in professional experience, academic research, policy engagement, or a combination of these.
Are interested in continuing to develop this expertise.
Know where to look and what to look for in order to ensure a comprehensive coverage of assigned topics, technologies, processes, etc.
Can identify major developments, policy controversies, key debates, and emerging trends in the digital governance landscape, and cover them accurately and impartially.

This means combining subject expertise with editorial judgement, policy awareness, and a strong sense of knowledge curation.

Applicants must also have:

Availability to contribute on a regular basis. The fellowship is conducted online, with an expected commitment of at least 8 hours per week.
Strong analytical and writing skills in English.
Basic skills in using web and social media, as well as familiarity with generative AI tools.

What we offer

Digital Watch Knowledge Fellows will benefit from:

Onboarding and guidance on Digital Watch’s editorial and curation approach.
Training on observatory workflows and digital/AI tools.
Remuneration.
Visibility for their work among DW users (diplomatic communities in Geneva and other diplomatic centres, professionals from across all stakeholder groups dealing with digital topics, etc.)
Opportunities to promote their digital governance-related research through DW and Diplo networks.
Membership in a global community of experts and professionals working on digital governance.

Fellows are engaged on a consultancy/fee basis; The role does not constitute employment with DiploFoundation.

How to apply

Interested applicants are invited to complete the application form.
Application deadline: 31 May 2026

Advertisement, Poster, Art, Outdoors, Text, Nature

Weekly #260 Mission, money, and the future of OpenAI

24 – 30 April 2026

Note to our readers: This issue comes to your inbox on a Thursday, rather than tomorrow, 1 May, in observance of Labour Day. Expect the next issue next Friday, as is customary.

HIGHLIGHT OF THE WEEK

Mission, money, and the future of OpenAI

For the second week in a row, technocrats take centre stage in our Weekly newsletter. This time, we’re spotlighting a billionaire row: A courtroom battle between Elon Musk and Sam Altman is putting the origins and future of OpenAI under scrutiny.

At the centre of the dispute is a fundamental question: was OpenAI meant to remain a nonprofit serving humanity, or was a shift toward a profit-driven model always part of the plan?

Musk, a cofounder, argues he was misled. He claims that OpenAI’s leadership abandoned its original mission and pivoted toward commercialisation, particularly through partnerships and products like ChatGPT. His lawsuit seeks sweeping remedies: removing Altman and president Greg Brockman, forcing structural changes to OpenAI’s governance, and potentially awarding up to $150 billion in damages to its nonprofit arm.

OpenAI, backed by Microsoft, rejects this narrative. Its legal team frames the case as a competitive dispute—arguing that Musk raised objections only after OpenAI’s success and the emergence of rival efforts, such as his own AI venture. In court, both sides are leaning heavily on early emails, funding discussions, and conflicting interpretations of what ‘open’ and ‘nonprofit’ were supposed to mean in practice.

Book, Comics, Publication, Person, Clothing, Shorts, Face, Head, Footwear, Shoe

The big (business) picture. This case could redefine the nonprofit–for-profit hybrid model that underpins much of today’s AI ecosystem. OpenAI’s structure—a nonprofit overseeing a capped-profit entity—has been widely copied or studied. If the court rules that such a transition violated founding principles, it could force a rethink across the industry, especially for organisations balancing public-interest missions with the massive capital demands of AI development.

Second, the trial may set a precedent for AI governance and accountability. Musk’s argument hinges on the idea that AI labs developing potentially transformative—or risky—technologies should be bound by enforceable commitments to the public good. If courts start treating these commitments as legally binding rather than aspirational, companies could face stricter scrutiny over how they deploy and monetise AI.

Third, there are implications for competition in AI markets. OpenAI’s partnerships, particularly with major tech players, have already raised questions about the concentration of power. A ruling that forces structural separation could reshape the competitive landscape.

It does bear saying that Musk’s xAI has filed for an initial public offering. OpenAI is rumoured to be considering an IPO itself, slated for Q4 2026 to mid-to-late 2027.

The user’s POV. If OpenAI is forced to prioritise its nonprofit mission more strictly, users might see greater transparency—for example, more openness about how models are trained, how decisions are made, or how risks are managed. On the other hand, limiting commercial incentives could slow down development or reduce the scale of investment, potentially affecting how quickly tools improve.

If the current model is upheld, it will underline that market logic and commercial interest will drive AI development. In practical terms, users could face more tiered access, stronger platform lock-in, and less visibility into how systems operate.

Beyond that, if the spat amuses you, The Verge has reporters in the courtroom offering coverage and witty commentary.

IN OTHER NEWS LAST WEEK

This week in AI governance

The USA. Washington is quietly reversing course on its standoff with Anthropic. The White House is drafting executive guidance that would allow federal agencies to work with Anthropic again, despite the company previously being labelled a supply-chain risk by the Pentagon. The shift reflects internal fractures: while parts of the defence establishment remain wary, others see excluding frontier models like Mythos as strategically costly.

The EU-USA critical minerals alliance for the technological future

The EU and the USA have launched a coordinated framework to strengthen resilience in critical minerals supply chains, combining a strategic Memorandum of Understanding (MoU) with an Action Plan.

The MoU establishes a broad strategic partnership covering the entire critical minerals value chain—from exploration and extraction to processing, recycling, and recovery. It frames critical minerals as strategic assets underpinning defence readiness, technological development, and economic resilience. The partnership aims to secure diversified and sustainable supply chains through joint project development in the EU, US, and third countries, supported by coordinated investment tools, risk reduction mechanisms, and improved business linkages.

Beyond supply security, the MoU introduces cooperation on market governance and resilience tools. This includes addressing non-market practices and export restrictions, promoting standards-based and transparent markets, improving permitting processes, coordinating on stockpiling and crisis response, and strengthening oversight of strategic asset sales. It also expands cooperation on innovation, recycling, geological mapping, and investment coordination. The agreement is explicitly non-binding, relying on domestic implementation and voluntary coordination.

The Action Plan operationalises these commitments by outlining steps toward a potential plurilateral trade initiative with like-minded partners. It explores coordinated trade instruments such as border-adjusted price floors, standards-based markets, price gap subsidies, and offtake agreements, initially focused on selected minerals. It also proposes harmonised standards, investment screening coordination, joint R&D, stockpiling cooperation, and rapid response mechanisms to supply disruptions. Implementation is led by USTR and DG TRADE, with links to broader multilateral efforts such as the G7.

Why does it matter? This initiative reflects ever-intensifying geopolitical competition over control of critical minerals, which are essential inputs for semiconductors, batteries, defence systems, and clean energy technologies. Supply chains are currently highly concentrated, particularly in processing and refining stages, creating strategic vulnerabilities for both the EU and the USA. The countries say it themselves: By aligning trade tools, standards, and investment screening, the EU and USA aim to safeguard their technological future (including energy, automotive, and electronics sectors), defence readiness, and economic resilience against external disruptions.

Europe’s growing age verification push for platform use

The European Commission has urged member states to rapidly roll out an EU age-verification app that allows users to prove they meet minimum age requirements without revealing personal data such as identity or exact date of birth. The system is designed to integrate with national digital identity wallets and can either operate as a standalone application or be embedded into existing e-ID infrastructure

This initiative is part of a broader EU enforcement effort under the Digital Services Act (DSA), which requires platforms to take stronger measures to protect children online. The Commission has also recently taken preliminary action against Meta, finding that Facebook and Instagram have not effectively prevented users under 13 from accessing their services, largely because age checks can be bypassed with false birthdates and weak verification systems.

At the same time, several European countries are moving toward stricter national rules that go beyond platform compliance. Norway has announced plans to introduce a ban on social media use for children under 16, placing responsibility for age verification on technology companies. Greece is considering measures that would restrict anonymity online and strengthen digital identity requirements. Under the plan, social media platforms would, from 2027, be required to block access for users under 15 using age verification systems rather than self-declared age data.

Australia reshapes news bargaining rules

Australia’s government has proposed a new Media Bargaining Incentive designed to force large digital platforms to financially support local journalism—or pay a levy.

Under the plan, tech companies with significant Australian revenue (over $250 million annually) would face a charge of up to 2.25% of their Australian revenue if they do not reach commercial agreements with at least four news organisations. The revenue collected would be redistributed to media outlets, with allocations linked partly to newsroom staffing levels.

These agreements would be “super-deductible”, meaning firms could offset up to 150% of their value (or 170% for smaller publishers) against the levy. In practice, this makes negotiating with media outlets cheaper than paying the tax itself.

The government proposes the measure as a correction to an imbalance in the digital economy. Communications Minister Anika Wells argued that large platforms benefit directly from journalism flowing through their feeds and should therefore contribute to its production, especially as news consumption shifts overwhelmingly to social media.

The reaction from Big Tech has been sharp. Meta dismissed the measure as a ‘government-mandated transfer of wealth’, arguing that news organisations voluntarily publish content on its platforms because they derive value from it. It also warned that the scheme resembles a digital services tax. Google also rejected the policy, pointing to its existing commercial deals with more than 90 Australian news businesses and arguing that the proposal misunderstands how the advertising market and news consumption have evolved. Both companies also criticised the policy’s selective scope, which excludes major platforms such as Microsoft, Snapchat, and OpenAI.

Australian media organisations, by contrast, strongly support the move. In a joint statement, outlets including the ABC, News Corp Australasia, Nine, SBS, and others described the proposal as a critical step to ensuring the sustainability of journalism.

What’s next? The draft legislation will now enter consultation, with lobbying from both tech firms and media organisations expected to intensify as the details are finalised. Consultation on draft legislation is open until 18 May 2026.

View all updates

LOOKING AHEAD

It will be a busy week in Geneva as the Geneva Cyber Week 2026 unfolds. Organised by the UN Institute for Disarmament Research (UNIDIR) and the Swiss Federal Department of Foreign Affairs (FDFA) under the overarching theme ‘Advancing Global Cooperation in Cyberspace’. Discussions will cover topics such as cyber norms and international cooperation, AI governance and regulation, critical infrastructure protection, cyber capacity building, incident response, and the security implications of emerging technologies, including artificial intelligence and quantum computing. Today (30 April) is the last day to register for the event.

As part of Geneva Cyber Week, UNIDI will organise the Cyber Stability Conference 2026, on 4–5 May in Geneva and online, bringing together governments, international organisations, industry, academia, and civil society to discuss ICT security and cyber governance. Under the theme “Cyber governance in an era of technological revolution: Past lessons, present realities and future frontiers,” discussions will explore how international cyber stability frameworks are adapting to rapid technological change, including AI and quantum computing, while reflecting on lessons from past cyber diplomacy processes and current security challenges.

Meanwhile, RightsCon 2026, which was scheduled to kick off in Lusaka, Zambia, on 5 May, will not proceed either in Lusaka or online. The conference has been deferred to a later date, the Zambian government has stated.

READING CORNER

writers ai writing style authorship debate

When AI writes the rules: How to avoid fake laws governing real life – Diplo

AI systems are increasingly capable of producing legal language and rules that look authoritative, including cases where outputs have echoed or fabricated legal references, as highlighted in South Africa. The real question, writes Jovan Kurbalija, is how societies can distinguish between useful AI assistance and ‘fake laws’ and why human institutions must remain the final gatekeepers of legitimacy and enforcement.

Open-Weight AI takes root – Diplo

In this blog, Slobodan Kovrlija examines how open-weight AI is empowering emerging economies to build sovereign agricultural and health tools, from Kenya’s crop diagnostics to Zambia’s maternal care.

Weekly #259 The ‘Technological Republic’ tech oligarchs imagine

17 – 24 April 2026

HIGHLIGHT OF THE WEEK

The ‘Technological Republic’ tech oligarchs imagine

Last week, the Technological Republic Manifesto by Palantir’s founder, Alex Karp, triggered an avalanche of comments and criticism as he challenged many pillars of our society, from equality and inclusion to security and democracy.

In 22 points extracted from his book, ‘Technological Republic’, Alex Karp, Palantir’s CEO, mixes national security, techno optimism, and democracy scepticism.

Palantir has already been at the centre of controversies around the use of its security products – Gotham, Foundry, and Maven in the Gaza war and by the US security apparatus on anti-migration and criminal activities.

The new Manifesto added to the controversy as the company moved from the business realm to the ideological and political realms.

Cas Muddle labelled Manifesto “Technofascism pure!” while Yanis Varoufakis said if Evil could tweet, this is what it would!

The backlash has been especially strong in the UK, where Palantir already holds major public contracts. Critics, including MPs and campaign groups, argue that the company’s ideology sits uneasily with its presence in sensitive parts of the state, from health data to policing and defence. Palantir, by contrast, says its systems improve efficiency, resilience, and public services.

Why does it matter? Palantir’s Manifesto put in a sharper and blunter way the growing power of tech companies in shaping society both nationally and internationally. Modern societies will have to get right what type of legal and policy order is needed and how to deal with the growing power of tech companies and their leaders.

IN OTHER NEWS LAST WEEK

This week in AI governance

EU’s defence cloud reliance raises ‘kill switch’ fears.

A new report says most of the EU defence agencies remain heavily dependent on US cloud providers, exposing critical systems to the risk of a foreign ‘kill switch’ and sharpening concerns over Europe’s digital sovereignty.

According to the findings, 23 of 28 countries studied rely on US tech for defence functions, with 16 assessed as high risk, prompting renewed debate over whether sensitive public infrastructure, including security and defence systems, should move faster toward sovereign or air-gapped alternatives.

France vs X, a transatlantic showdown

France’s criminal investigation into X has evolved into a transatlantic dispute over platform governance and state authority.

How did it all begin? The case began with a French probe into whether the platform enabled the spread of child sexual abuse material, AI-generated deepfakes, Holocaust denial content, and other harmful or unlawful material, and later intensified with a search of X’s Paris offices and summonses for Elon Musk and former X chief Linda Yaccarino to give voluntary interviews – a request Musk appears to have refused by not showing up.

And then. The confrontation widened when reports emerged that the US Justice Department had declined to assist the French inquiry, arguing that the case risked crossing into the regulation of protected speech and that it would unfairly target a US company. French authorities, however, have framed the matter as a legitimate enforcement action under national law.

Australia targets also games in child safety crackdown

Seen in context. This is part of a broader tightening of enforcement around Australia’s under-16 social media rules, which took effect on 10 December 2025 and require age-restricted platforms to take reasonable steps to prevent underage children from creating and holding accounts. Yet regulators say compliance remains uneven: in March, eSafety flagged big concerns about Facebook, Instagram, Snapchat, TikTok and YouTube, warning that many children could still access platforms by simply self-declaring they were older than 16.

Microsoft bets A$25 billions on Australia’s AI future

Microsoft has announced a A$25 billion investment in Australia by 2029, its largest in the country, to expand local AI and cloud infrastructure, strengthen cybersecurity, and train three million Australians in workforce-ready AI skills.

The plan will increase Azure AI supercomputing capacity, expand Microsoft’s Australian cloud footprint by more than 140%, and deepen cooperation with the Australian Government, including the Australian AI Safety Institute and the Microsoft–Australian Signals Directorate Cyber Shield.

Framed as support for Australia’s National AI Plan, the package links AI growth with cyber resilience, digital sovereignty, responsible deployment, and broader access to skills across schools, nonprofits, workers, government, and industry.

UK fortifying child safety online with new powers

The UK’s Children’s Wellbeing and Schools Bill would reportedly expand ministers’ powers to shape how online services protect children, including by restricting access to risky platforms, features, or functions and by targeting design elements such as contact settings, live communication, location visibility, and time spent online.

The draft would also bring Ofcom into a stronger advisory role, introduce a six-month timeline for regulations or a progress update, and give ministers new authority over children’s data consent, age assurance, and enforcement.

Why does it matter? Taken together, the amendments point to a more interventionist and fine-grained model of child online safety, focused not only on harmful content but also on the design and governance of children’s digital environments. The regulatory package remains unsettled for now, with Parliament still negotiating key provisions and no final law yet in place.

View all updates

LAST WEEK IN GENEVA

Shaping Switzerland’s AI Summit Strategy

29th session of the CSTD

The 29th session of the Commission on Science and Technology for Development (CSTD) is ending today (Friday). The programme addressed the priority theme of ‘Science, Technology and Innovation in the Age of Artificial Intelligence’ and also reviewed progress in implementing and following up to the outcomes of the World Summit on the Information Society at regional and international levels. We’ll have more on the outcomes next week.

READING CORNER

Claude Mythos and the myth of AI control – Diplo

Anthropic’s Mythos model is a cyber-offensive AI built to probe critical infrastructure. Why does this reality expose the flaws in current AI governance?

Claude Mythos Preview sets new benchmark for AI capability and raises governance questions

Anthropic’s Claude Mythos Preview is its most capable model to date, withheld from public release and made available only to a closed partner network amid concerns about its cybersecurity capabilities and governance implications.

Weekly #258 European firms build a digital ‘backup generator’

10 – 17 April 2026

HIGHLIGHT OF THE WEEK

European firms build a digital ‘backup generator’

A group of European firms has unveiled what it effectively describes as a digital ‘backup generator’—a full-stack recovery system designed to keep critical services running in the event of access to foreign technology providers being disrupted.

Developed by Cubbit, Elemento, SUSE, and StorPool Storage, the so-called ‘Disaster Recovery Pack’ was launched on 15 April in Berlin at the European Data Summit hosted by the Konrad-Adenauer-Foundation.

Box, machine, Wheel, Cardboard, Carton, Package, Package Delivery, Person, Car, Transportation, Vehicle

The Pack bundles together storage, compute, orchestration, networking, identity, observability, and management into a pre-integrated, deployable system. Organisations can use the system to identify critical services, build a sovereign recovery setup, and shift key operations to a fully European stack in the event of disruption.

The aim is not to replace existing infrastructure outright, but to give organisations a ready-to-activate fallback environment—one that can be tested in advance and scaled progressively across workloads.

While the notion of a foreign vendor ‘kill switch’ remains contested, the underlying concern—loss of access to critical services due to external legal, political, or commercial decisions—has gained traction across European policy circles.

That concern is reinforced by market structure. US firms, including Google, Amazon Web Services, and Microsoft, continue to dominate Europe’s cloud ecosystem, while payment systems and software layers remain similarly concentrated.

In this context, resilience is increasingly framed not as full technological independence, but as the ability to withstand disruption without systemic failure.

Why does it matter? Initiatives like the Disaster Recovery Pack could form the backbone of a more resilient European digital ecosystem—one designed not to eliminate dependencies, but to manage them on Europe’s own terms.

IN OTHER NEWS LAST WEEK

This week in AI governance

South Africa. South Africa has unveiled a draft national AI policy proposing new institutions — including a National AI Commission, an AI Ethics Board and a regulatory authority — alongside incentives such as tax breaks and grants to boost local innovation. The plan aims to position the country as a continental AI leader while addressing governance, infrastructure and data sovereignty concerns.

Russia. Russia is advancing a draft AI regulatory framework that would formalise oversight of AI development and deployment, aligning with broader efforts to strengthen digital sovereignty and state control over emerging technologies. The proposals focus on risk management, national standards and reducing dependence on foreign AI systems, while supporting domestic innovation. The move fits into Moscow’s wider strategy of tightening control over digital infrastructure and cross-border data flows.

UNESCO — Latin America & Caribbean. UNESCO has launched a regional AI in Education Observatory for Latin America and the Caribbean, designed to support evidence-based policymaking and track the impact of AI on education systems. The initiative aims to build capacity, share best practices and guide responsible integration of AI tools in schools and learning environments.

Ghana. The Ghanaian Ministry of Communication, Digital Technology and Innovations has launched a public-sector AI capacity development programme in collaboration with the Government of Japan and the United Nations Development Programme. The programme is designed to equip public officials with knowledge of AI and its applications in governance. It focuses on improving decision-making and service delivery, drawing on experience from the UN and Japan.

EU develops age verification app

The app is open source and available for both public and private implementation, but is subject to common technical and privacy requirements. The Commission plans to establish an EU-level coordination mechanism to oversee rollout, accreditation, and cross-border usability.

This technical and regulatory push is unfolding alongside political coordination among member states. Several member states are already preparing to integrate the app into national digital identity wallets, with France, Denmark, Greece, Italy, Spain, Cyprus and Ireland cited as front-runners. French President Emmanuel Macron is convening EU leaders, including Spanish Prime Minister Pedro Sanchez and representatives of Italy, the Netherlands and Ireland, to align national approaches to restricting minors’ access to social media and to press for faster EU-level action.

Yes, but. The rollout is already facing scrutiny. Shortly after Ursula von der Leyen described the app as technically ready and privacy-preserving, a security researcher claimed its protections could be bypassed in minutes. The critique points to structural design issues rather than isolated bugs. Reported weaknesses include locally stored authentication data that can be reset or modified, allowing users to bypass PIN protections, disable biometric checks, and reset rate-limiting mechanisms by editing configuration files. This effectively enables the reuse of verified identity data under altered access controls.

The criticism has triggered broader concerns among developers about the app’s architecture, including why secure hardware features were not used, and whether elements like expiring age credentials are logically necessary.

UK threatens jail for tech executives over non-consensual sex images removal

The move builds on existing obligations that already require platforms to take down such material within strict timeframes or face significant penalties, including fines of up to 10% of global turnover or even service blocking.

Why does it matter? The latest step goes further: instead of relying solely on corporate sanctions, it introduces personal criminal accountability at the executive level. This type of liability is likely to accelerate compliance in ways that financial penalties alone have not, and may serve as an example to other jurisdictions.

Zooming out. The policy is part of a broader tightening of the UK’s online safety framework, driven by persistent concerns over revenge porn and the rapid proliferation of AI-generated intimate imagery.

EU blocks Meta’s WhatsApp third-party AI access changes with interim antitrust measures

Previously, Meta decided to reinstate access to WhatsApp for third-party AI assistants for a fee. However, the Commission has preliminarily found that these measures remain anticompetitive and has now issued interim measures to prevent these policy changes from causing serious harm on the market.

The interim measures would stay in effect until the Commission concludes its investigation and issues a final decision on Meta’s conduct.

View all updates

LOOKING AHEAD

The 29th session of the Commission on Science and Technology for Development (CSTD) is scheduled to take place from 20 to 24 April 2026 at the Palais des Nations in Geneva, Switzerland.

For its 29th session, the programme will address the priority theme of ‘Science, Technology and Innovation in the Age of Artificial Intelligence’ and will also review progress in implementing and following up to the outcomes of the World Summit on the Information Society at regional and international levels.

The session will include presentations on technical cooperation activities and the work of the multistakeholder Working Group on Data Governance, as relevant to development objectives. Participation is expected from representatives of national governments, international organisations, civil society and the private sector.

READING CORNER

The gap between AI rules and AI reality – Diplo

An analysis of why the EU AI Act’s high-risk obligations are delayed by 16 months and how US federal intervention is dismantling state-level AI safety laws, creating a global governance vacuum.

Weekly #257 AI meets cybersecurity as project Glasswing takes flight

3 – 9 April 2026

HIGHLIGHT OF THE WEEK

AI meets cybersecurity as project Glasswing takes flight

This week, a veritable who’s who of tech—Amazon, Apple, Google, Microsoft, NVIDIA, and a dozen other giants—joined the Anthropic-led cybersecurity project Glasswing.

The launch partners will use Antropic’s unreleased Claude Mythos Preview as part of their defensive security work, a tool the company claims can already identify software vulnerabilities at a level surpassing that of most human experts.

The premise is straightforward and difficult to dispute: If AI systems can find and exploit vulnerabilities at scale, then those same capabilities should be deployed defensively, before less scrupulous actors gain access. Anthropic frames this as a narrow window of opportunity. Mythos Preview, it argues, has already uncovered thousands of high-severity vulnerabilities across major operating systems and browsers—an assertion that, if accurate, signals a step-change in the automation of software exploitation.

Yet the announcement also raises questions that go beyond the promises.

There is the question of verification. Claims that a model can ‘surpass all but the most skilled humans’ at vulnerability discovery are inherently difficult to evaluate externally, particularly when the system itself is not publicly available.

A systemic issue that could arise is coordination. If AI accelerates the rate at which vulnerabilities are found, it may also overwhelm remediation efforts, effectively creating a bottleneck.

The model remains unreleased, accessible only to a curated group of partners and selected infrastructure maintainers. This controlled access concentrates a powerful capability in the hands of a small set of actors. Smaller vendors, public institutions, and under-resourced open-source projects may benefit indirectly from disclosed fixes, but they are unlikely to operate on equal footing.

It is also worth noting that all core partners in Project Glasswing—from Amazon Web Services and Google to Microsoft, Apple, and Cisco—are headquartered in the United States. That matters, because access to the most sensitive capability—the model itself—appears tightly governed and selectively distributed. Even if non-US entities participate, they are unlikely to do so on equal terms. It reflects where frontier AI development and much of the global cybersecurity industry are currently anchored, but it also reinforces the geopolitical framing that increasingly surrounds these technologies.

That said, it would be misleading to see this as purely exclusionary. If the initiative results in patched vulnerabilities, improved open-source security, and shared findings, its effects will be globally distributed—whether or not governance is.

IN OTHER NEWS LAST WEEK

This week in AI governance

OpenAI has released a policy document entitled ‘Industrial Policy for the Intelligence Age: Ideas to Keep People First’. The document argues that while superintelligence promises extraordinary benefits, it also carries serious risks: job displacement, misuse by bad actors, loss of human control, and concentration of power and wealth. The proposals are organized into two sections. First, building an open economy: giving workers a voice in AI deployment, treating AI access as a fundamental right, creating a ‘Public Wealth Fund’ to give citizens direct stakes in AI growth, converting efficiency gains into shorter workweeks, and building adaptive safety nets that trigger automatically when disruption occurs. Second, building a resilient society: developing containment playbooks for dangerous AI, create verifiable trust stacks for content, strengthen independent auditing of frontier models, mandate incident reporting, and build international information-sharing networks.

The EU. If you want to let European lawmakers know what you think of the implementation of the bloc’s AI Act, there is still a bit of time. The feedback period on the draft Implementing Regulation related to the oversight of general-purpose AI models under Regulation (EU) 2024/1689 (the EU Act) will remain open until tonight, 9 April (midnight).

US Supreme Court narrows ISP liability, sharpens focus on intent with AI implications

A unanimous US Supreme Court ruling this week has narrowed the circumstances under which an internet service provider (ISP) can be held liable for users’ copyright infringement.

Writing for the Court, Justice Clarence Thomas said an ISP is liable only if its service was designed for unlawful activity or if it actively induced infringement.

The decision could have implications beyond ISPs, particularly in the escalating copyright battle between publishers/authors and generative AI firms.

The key distinction raised is that broadband networks function as neutral conduits, whereas large language models are built specifically to produce fluent, human-like writing, including prose, poetry and dialogue, that can resemble the work of human authors. If a subscriber uses broadband to pirate a novel, the ISP did not build its network to enable that outcome, but an AI model prompted to write in a specific author’s style is designed to fulfil that request.

US agencies warn of cyber intrusions into critical infrastructure systems

A joint cybersecurity advisory issued by the Federal Bureau of Investigation, Cybersecurity and Infrastructure Security Agency, National Security Agency, and several sector-specific partners warns US organisations of an ongoing campaign by actors targeting industrial control systems across US critical infrastructure.

The activity focuses on internet-exposed operational technology (OT), particularly programmable logic controllers (PLCs), which are widely used to automate industrial processes in sectors such as energy, water and wastewater systems, and government services.

According to the advisory, the attackers are exploiting PLCs by leveraging their direct exposure to the internet. The attackers gain initial access by scanning for internet-facing PLCs and connecting through commonly used industrial communication ports. Once access is established, the actors interact with device project files and manipulate data displayed on human-machine interfaces (HMI) and supervisory control and data acquisition (SCADA) systems. This enables them to disrupt industrial processes in real time. In several confirmed cases, such intrusions have resulted in operational disruption and financial loss, underscoring the tangible, physical-world impact of these cyber operations.

The campaign appears to be part of a broader escalation in Iranian-linked cyber activity, likely tied to geopolitical tensions involving the USA and its allies. The advisory links the activity to previously identified advanced persistent threat (APT) groups associated with Iran’s Islamic Revolutionary Guard Corps (IRGC).

Greece sets ‘digital age of majority,’ moving to ban under-15s from social media

Platforms would be required to implement robust age verification mechanisms, including the re-verification of existing accounts, with oversight provided by national regulators such as the Hellenic Telecommunications and Post Commission (EETT).

The measure applies to social networking services where users create profiles, publish content, and interact publicly, while excluding private communication services.

The big picture. The proposal reflects an emerging policy pattern across Europe, where governments are increasingly willing to intervene more directly in platform access for minors. Athens is also seeking to elevate the issue at the European level. ‘Our goal is to push the European Union in this direction as well,’ Prime Minister Kyriakos Mitsotakis noted in a video about the measure posted on TikTok.

Brazil launches first national centre for assistive technology

The launch marks the first of several planned centres nationwide to expand access to these technologies.

Yes, but. The long-term impact will depend on sustained investment and the ability to scale these centres nationwide.

View all updates

UPCOMING EVENTS

WTO deadlock, AI boom: Unpacking MC14 and looking ahead

Diplo, the Digital Trade and Data Governance Hub, and the Geneva Internet Platform will co-organise a webinar on 14 April (next Tuesday) that unpacks digital trade-related developments in the 14th WTO Ministerial Conference in Yaoundé and looks ahead to their implications for the rapidly expanding AI economy. As digital trade rules take shape through multiple channels, understanding the intersection between trade policy and AI governance becomes increasingly urgent. The speakers will explore:

What to expect at the next General Council meeting in May and beyond

The main outcomes and sticking points from MC14

What the lapse of the e-commerce moratorium means — and what it does not mean

How the plurilateral JSI e-commerce agreement may shape digital trade going forward

The specific implications for AI development, including data flows, tariffs on digital services, and regulatory coherence

Registration for the event is open.

READING CORNER

European Commission EU AI Act amendments Digital Omnibus European AI Office

The implementation of the EU AI Act with a focus on general-purpose AI models

The European Union is progressing into the implementation phase of its Artificial Intelligence Act, with emerging obligations for providers of general-purpose AI models. Guidance from the European Commission and the AI Office outlines compliance expectations as the EU operationalises its risk-based AI governance framework.

Campaigns

HIGHLIGHT OF THE WEEK

IN OTHER NEWS LAST WEEK

LAST WEEK IN GENEVA

LOOKING AHEAD

READING CORNER

OPPORTUNITY

Rétrospective de avril 2026

Coup d’oeil : les événements qui ont fait la une en avril

TECHOLOGIES

Infrastructure

SÉCURITÉ

ÉCONOMIE

DROITS DE L’HOMME

JURIDIQUE

SOCIOCULTUREL

Ce mois-ci dans la gouvernance de l’IA

Plans et initiatives nationaux

Souveraineté

Partenariats

Juridique

Sécurité et sûreté

Gouvernance des contenus

Développement

Quand l’IA écrit les règles : comment éviter que des lois fictives régissent la réalité

Concevoir notre avenir

Claude Mythos Preview établit une nouvelle référence en matière de capacités de l’IA et soulève des questions de gouvernance

Capacités en matière de cybersécurité

Projet Glasswing

Dimensions géopolitiques

La marche à suivre

Le mois dernier à Genève

29e session de la CSTD

Élaboration de la stratégie suisse pour le Sommet sur l’IA

Semaine de la cybersécurité de Genève 2026

Opportunité : Devenez Chargé de recherche

Qu’est-ce que le Digital Watch Observatory ?

Quel est le rôle d’un chargé de recherches ?

Qui peut postuler ?

Ce que nous offrons

Comment postuler

HIGHLIGHT OF THE WEEK

IN OTHER NEWS LAST WEEK

LAST WEEK IN GENEVA

READING CORNER

OPPORTUNITY

HIGHLIGHT OF THE WEEK

IN OTHER NEWS LAST WEEK

LAST WEEK IN GENEVA

READING CORNER

OPPORTUNITY

Opportunity: Become a Knowledge Fellow

What is the Digital Watch Observatory?

What is the role of a Knowledge Fellow?

Who should apply?

What we offer

How to apply

April 2026 in retrospect

Snapshot: The developments that made waves in April

Technologies

Infrastructure

Security

Economic

Human rights

Legal

Sociocultural

AI governance in April

National plans and initiatives

Sovereignty

Partnerships

Legal

Safety and security

Content governance

Development

When AI writes the rules: How to avoid fake laws governing real life

Designing our future

Claude Mythos Preview sets new benchmark for AI capability and raises governance questions

Cybersecurity capabilities

Project Glasswing

Geopolitical dimensions