The AI Safety Institutes Network is an international coalition of national AI safety institutes established in 2024. The network brings together government-backed research bodies focused on evaluating and mitigating risks from advanced Artificial intelligence systems, particularly Large language models and Generative AI technologies.
Member institutes collaborate on shared research agendas, benchmark development, and information exchange. The network emerged from a series of international summits on AI safety, reflecting growing government interest in regulating frontier AI development.
Formation and Members
The network was launched following the AI Safety Summit held at Bletchley Park in November 2023å¾’ . Initial founding members included the AI Safety Institute from the United Kingdom and the U.S. AI Safety Institute, established under the National Institute of Standards and Technology. By late 2024, the network had expanded to include institutes from the European Union, Japan, Canada, Singapore, and several other nations. Each member institute operates independently under its national mandate but participates in joint technical workshops and shared evaluation exercises.
Core Activities
The network's primary functions include developing common evaluation standards for frontier AI models, sharing safety research findings, and coordinating policy recommendations. Member institutes collaborate on red-team testing of models from organizations such as OpenAI, Anthropic, and Google DeepMind. Joint projects have produced public benchmarks for assessing risks in areas like cyber ability, biological misuse, and autonomous decision-making. The network also maintains a shared repository of incident reports and safety incidents involving deployed AI systems.
Governance and Structure
The coalition operates without a permanent secretariat initially, instead rotating coordination duties among member institutes. The network holds semi-annual plenary meetings and maintains working groups focused on specific topics such as model evaluations, alignment research, and international standards alignment with bodies like the OECD and the Global Partnership on AI. Member institutes contribute staff and resources to joint projects on a voluntary basis.
Policy Impact
The AI Safety Institutes Network has influenced national regulations and industry practices. Its evaluation frameworks have been referenced in the European Union's AI Act implementation guidance and in the White House Executive Order on Safe, Secure, and Trustworthy Development of AI. Some member governments have cited network findings when requiring safety assessments from AI developers before model release. The network's public reports have informed corporate risk-management approaches at major AI labs and cloud providers such as Microsoft Azure and Google Cloud.
Challenges and Future Directions
As of 2025, the network faces challenges related to differing national priorities, intellectual property concerns regarding proprietary model evaluations, and the pace of AI advancement outstripping assessment capabilities. Proposed next steps include expanding membership to additional countries, developing standardized incident reporting protocols, and establishing a permanent funding mechanism. The network continues to emphasize international cooperation as essential for managing global AI risks, while acknowledging tensions between safety enforcement and technological competitiveness.
Reception and Criticism
Industry observers have generally welcomed the network as a coordinating mechanism, though some researchers argue that evaluation standards remain too narrowly focused on technical failures rather than societal impacts. Civil society groups have called for greater transparency and civil society participation in network activities. The coalition's relationship with frontier AI companies has evolved from adversarial assessments to collaborative evaluations, mirrored by similar dynamics at individual national institutes.
See Also
- Machine learning
- Neural network
- AI safety
- international-ai-governance
References
Sources: Bletchley Declaration 2023; official announcements from member institute websites; coverage in major technology and policy journals through 2025.