Artificial intelligence Archives | Page 16 of 667

United Kingdom and Australia tighten alliance on AI security risks

The United Kingdom and Australia are deepening cooperation on AI security through a new partnership between the UK AI Security Institute and the Australian AI Safety Institute.

Under a Memorandum of Understanding, the two institutes will share information on frontier AI capabilities, collaborate on AI evaluation practices and exchange research findings. The UK government said the partnership will focus partly on how advanced AI systems could be used in cyberattacks, as well as how they can strengthen defensive capabilities.

The agreement will also support staff exchanges between the two institutes, strengthening day-to-day collaboration. UK officials said the partnership reflects the need for trusted international cooperation as AI systems evolve quickly and create new security and safety risks.

The UK’s AI Minister Kanishka Narayan is expected to sign the agreement with Australia’s Assistant Minister for Science, Technology and the Digital Economy, Andrew Charlton, during a meeting in Canberra. Narayan said no country can address fast-moving AI risks alone, particularly in cybersecurity.

The announcement follows research from the UK AI Security Institute showing that advanced AI systems are rapidly improving their ability to carry out complex cyberattacks, creating opportunities for both attackers and defenders. The UK said the institute’s frontier AI research continues to inform policymaking to protect businesses, critical infrastructure, and the public.

Why does it matter?

The partnership shows how AI security is becoming a matter of international coordination, especially as frontier models develop stronger cyber capabilities. By sharing research, evaluation methods and staff expertise, the UK and Australia are trying to reduce blind spots in oversight and develop more consistent approaches to testing fast-moving AI systems.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our chatbot!

EU consultation closes on AI energy measurement

The European Commission has moved forward with work on measuring the energy consumption and emissions of AI models and systems, as part of preparations for a possible AI energy measurement framework under the EU AI Act.

The targeted consultation forms part of a Commission-procured study on measuring and promoting energy-efficient and low-emission AI in the European Union. Responses will help refine the study, contribute to a measurement framework for the AI Act’s energy-related objectives and support the design of a potential AI energy and emissions label.

The process focuses on how to measure energy use across the AI lifecycle, including development and training, as well as operational use and inference. The Commission says a comprehensive picture of AI’s energy efficiency and carbon footprint requires data on computational resources, electricity consumption and hardware details.

Under Annex XI of the AI Act, providers of general-purpose AI models must document known or estimated energy consumption as part of their technical documentation obligations. The consultation, therefore, targets developers and deployers of general-purpose AI models and AI systems, as well as component and service suppliers.

Stakeholders were asked about the accessibility of data needed to assess AI energy consumption and emissions, as well as the suitability of different AI performance indicators. The Commission said the aim is to develop a robust and practical industry-informed framework for measuring AI energy consumption and efficiency.

The AI Office will publish a summary of the consultation results based on aggregated data, with respondents not directly quoted.

Why does it matter?

AI’s growing energy demand is becoming a regulatory and environmental policy concern, especially as general-purpose AI models require substantial computing resources for training and inference. A common EU framework for measuring AI energy use and emissions could make environmental impacts more visible, support future transparency obligations and help compare systems more consistently. A possible AI energy and emissions label would also push sustainability into AI governance alongside safety, transparency and accountability.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

European Commission marks 10 years of GDPR

The European Commission has marked ten years since the General Data Protection Regulation (GDPR) entered into force across the European Union.

The GDPR entered into force on 24 May 2016 and established a common data protection framework across EU member states, and introduced rules governing the collection and processing of personal data. According to the European Commission, the regulation strengthened individuals’ rights regarding how personal data is collected, processed, corrected, deleted, and shared.

The framework applies to organisations ranging from small businesses to multinational technology companies. Authorities across the EU have also issued significant penalties in cases involving non-compliance with the regulation.

The GDPR has influenced privacy and data protection discussions internationally and contributed to wider adoption of similar regulatory approaches.

The Commission linked the GDPR to broader EU digital regulation efforts, including the Digital Services Act, the Digital Markets Act, and the AI Act. According to the Commission, these measures address issues including platform accountability, competition, and AI governance.

The Commission also referenced online child protection initiatives, including work on age verification and cyberbullying prevention. It said the EU’s approach reflects the principle that the online world should serve people.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

Canada launches AI learning initiative for federal public servants

Canada’s School of Public Service is organising the Learning Week on Artificial Intelligence, an initiative aimed at strengthening AI understanding across the federal public service.

The programme is linked to the Government of Canada’s AI Strategy for the Federal Public Service 2025–2027.

The AI learning programme is open to public servants at all levels and across the country. The initiative includes live events, virtual sessions, self-paced learning tools, and practical demonstrations related to AI technologies.

According to organisers, the programme aims to improve awareness of AI-related opportunities, challenges, and skills within the public service.

The initiative also aligns with broader public service priorities involving digital transformation, productivity, and process modernisation.

Sessions will examine potential applications of AI in areas including policy development, service delivery, and internal administrative functions.

The programme is intended to support responsible AI adoption and prepare public servants in Canada for organisational and operational changes linked to AI technologies.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

Anthropic says AI system identified thousands of critical software flaws

Anthropic has published an update on Project Glasswing, a cybersecurity initiative focused on identifying software vulnerabilities using AI systems.

According to Anthropic, partner organisations used Claude Mythos Preview to identify thousands of high- and critical-severity vulnerabilities across software platforms and infrastructure systems.

The company said the initiative demonstrated how AI systems are increasing the speed and scale of vulnerability discovery processes. Anthropic reported that participating organisations observed substantial increases in software vulnerability detection capabilities during testing.

Evaluations cited by Anthropic suggested the system performed strongly in vulnerability identification and exploit-detection tasks compared with earlier AI cybersecurity models.

Anthropic also said the model analysed more than 1,000 open-source projects and identified vulnerabilities affecting widely used software components. The company highlighted a vulnerability identified in the open-source cryptography library wolfSSL as one example from the project.

According to Anthropic, the vulnerability was patched after disclosure.

Anthropic said AI-assisted vulnerability discovery may increasingly shift cybersecurity challenges toward verification, disclosure, and remediation processes. The company also said similar AI cybersecurity capabilities are likely to become more widely available across the industry.

Why does it matter?

The rapid growth of AI-driven cybersecurity is becoming increasingly important as AI is fundamentally changing the balance between cyber defence and cyber threats. Systems such as Anthropic’s Project Glasswing demonstrate that advanced AI models can identify software vulnerabilities at a speed far beyond traditional human-led security testing, potentially making critical infrastructure, financial systems, cloud platforms, and open-source software both safer and more exposed at the same time.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

UN DESA launches AI governance workshop for Africa and Asia-Pacific officials

The United Nations Department of Economic and Social Affairs (UN DESA) is organising a capacity-building workshop in Hangzhou, China, focused on AI governance and digital transformation for senior government officials from Africa and the Asia-Pacific region.

The activity is part of the project ‘Strengthening AI Capabilities and Governance for Sustainable Digital Transformation in Africa and Asia-Pacific’, implemented in partnership with the Government of China. The programme includes a five-day study tour examining policy, institutional, and technical approaches to AI governance and adoption.

According to the concept note, the initiative responds to gaps in national AI governance frameworks and implementation capacity in many countries. The programme references findings from the UN E-Government Survey 2024 and OECD data concerning AI regulation and national AI strategies.

The programme will draw on the UN E-Government Survey 2024 and its AI Addendum, as well as China’s experience in using AI to support micro, small, and medium-sized enterprises, poverty reduction, and inclusive growth.

Participants are expected to review AI governance structures, regulatory frameworks, and institutional coordination mechanisms. The agenda also includes briefings with relevant ministries or AI bodies, as well as visits to AI coordination or digital transformation institutions.

The agenda includes discussions on international cooperation, regulatory interoperability, public-private collaboration, and AI-related opportunities for small businesses.

According to the concept note, the workshop aims to strengthen institutional understanding of AI governance and support integration of digital policies into national development strategies. Participation is by invitation only and limited to around 20 attendees.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

EuroDIG 2026 to bring European internet governance voices to Brussels

EuroDIG 2026 will take place on 26 and 27 May at the Charlemagne Building of the European Commission in Brussels, bringing together Europe’s internet governance community for two days of discussions on the future of the digital environment.

The event will be hosted by EURid, the registry operator for the .eu domain name, with support from the European Commission, a longstanding institutional partner of EuroDIG. This year’s edition also marks 20 years of .eu domain, celebrating two decades of what organisers describe as a trusted European digital identity.

The overarching theme is ‘European Voices for the Future of the Internet – Celebrating 20 Years of .eu and the Beginning of a New Internet Governance Era’. Discussions are expected to address issues including openness, security, multistakeholder governance, and Europe’s digital policy priorities.

Over the past 18 years, EuroDIG has served as a European multistakeholder platform for discussions on internet governance and digital public policy. Outcomes from the discussions contribute to broader international internet governance processes, including the Internet Governance Forum.

Participants from government, civil society, academia, the technical community, business, and youth groups are expected to take part in the discussions. Sessions will address topics including AI, digital identity, information integrity, infrastructure resilience, digital sovereignty, and democracy online.

Digital Watch Observatory is following EuroDIG 2026 through a dedicated event page, featuring session information and reporting from Brussels.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

European Commission advances AI transparency code under EU AI Act

The European Commission’s AI Office has convened a new round of working group meetings and workshops on the forthcoming Code of Practice on Marking and Labelling of AI-Generated Content.

The discussions brought together providers of generative AI systems and models, technology companies, industry representatives, civil society organisations and academic experts. Feedback from the meetings will inform the third and final draft of the code, expected in early June.

The code is intended to support transparency obligations under the AI Act, including requirements linked to marking, labelling, disclosure and detectability of AI-generated content. It covers issues such as synthetic media, deepfakes and certain AI-generated text.

Working Group 1 focused on marking and detection obligations for providers, including a revised multi-layered approach, technical feasibility, benchmarking, compliance frameworks and possible third-party assessments. Industry participants raised concerns over compliance burdens, innovation and feasibility, while civil society and academic experts called for stronger safeguards in the public interest.

Working Group 2 examined disclosure obligations for deployers of generative AI systems, particularly deepfakes and certain AI-generated text. Discussions covered origin disclosure, user-facing labels, proportionality, governance measures, editorial control and the possible development of a uniform EU label.

Additional workshops explored how machine-readable marks, provenance data, visible labels, watermarking systems and an EU-wide icon could work together across the AI value chain. Participants also discussed coordination with other EU rules, including the Digital Services Act, while stressing the need to balance transparency, legal clarity, accessibility and innovation.

Why does it matter?

The code of practice will help determine how AI-generated content is marked, labelled and disclosed across the EU. Its development highlights the practical difficulty of turning transparency obligations into workable rules, particularly when regulators, companies and civil society disagree over technical feasibility, compliance costs, user experience and safeguards against deceptive synthetic media.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Grokipedia articles show selective political divergence from Wikipedia, research finds

A new study published in the Proceedings of the National Academy of Sciences examined structural and political differences between Wikipedia and Grokipedia, the AI-generated encyclopedia developed by xAI.

Researchers analysed 17,790 matched article pairs drawn from the 20,000 most-edited English-language Wikipedia entries. They found that Grokipedia articles are typically longer, more syntactically complex, and contain fewer references and hyperlinks per 1,000 words than their Wikipedia counterparts.

The study also identified a bimodal pattern across similarity measures, indicating that some Grokipedia entries closely resemble Wikipedia entries, while others diverge substantially in content and structure. Researchers said the findings suggest Grokipedia is not a fully independent alternative to Wikipedia, but often appears as an AI-mediated reconfiguration of Wikipedia content.

The analysis examined ideological differences by evaluating the political orientation of cited news media sources. Researchers found that divergence was concentrated primarily in politically and culturally sensitive topics, including religion, history, politics and literature.

Within those areas, Grokipedia articles showed a relative shift toward more right-leaning cited sources than Wikipedia. However, the study also noted that sources cited on both platforms remained predominantly left-leaning.

Researchers argued that Wikipedia’s human editorial processes make disputes, revisions and bias visible and contestable, while AI-generated systems may embed bias within more opaque automated workflows that are harder to scrutinise publicly.

The paper also raised broader concerns about the governance of AI-generated knowledge systems. Researchers warned that AI-generated encyclopedic content could shape future training datasets and automated information ecosystems, potentially reproducing or amplifying bias without sufficient transparency, accountability or human oversight.

Why does it matter?

The findings add to growing debates over AI-generated knowledge systems, political bias, citation quality and transparency. As generative AI increasingly produces reference and educational material, the key question is not only whether outputs are accurate, but whether their sources, editorial assumptions and revisions can be scrutinised. Grokipedia’s differences from Wikipedia show how automated knowledge systems may reshape information governance while making some forms of bias less visible.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

European Central Bank examines systemic risks linked to AI-driven finance

The European Central Bank (ECB) has published research examining how AI systems could affect financial stability as AI adoption expands across financial markets.

According to Eurosystem research, different AI architectures may produce significantly different market behaviours under similar economic conditions.

ECB simulations compared reinforcement learning systems with large language model-based agents operating in simulated financial environments. Researchers found that some reinforcement learning systems displayed coordinated responses resembling bank run dynamics in certain scenarios.

The report linked part of this behaviour to risk-avoidance patterns associated with prior negative outcomes.

Large language model-based systems showed lower coordination but more variable and unpredictable responses during periods of uncertainty. Despite receiving identical instructions, LLM-based agents frequently developed different assumptions about market behaviour, particularly during periods of moderate economic uncertainty.

ECB researchers noted that such inconsistency could create its own form of instability as AI-generated expectations diverge across financial markets.

The ECB suggested that wider AI adoption in finance may require updated risk management practices, investor awareness, and regulatory safeguards.

The research also highlighted the potential importance of existing market stabilisation measures, including circuit breakers and investor protection mechanisms.

Why does it matter?

AI is rapidly becoming embedded in trading, investment management, and financial decision-making across global markets, meaning flaws in AI behaviour could amplify systemic risks at unprecedented speed and scale.

The research signals that financial stability may increasingly depend not only on economic fundamentals and regulation, but also on the underlying architecture, coordination patterns, and predictability of the AI systems shaping market activity.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our chatbot!