France launches international initiative on AI safety and security
The Paris Call for AI Safety and Security brings governments, AI developers, researchers and civil society organisations together to strengthen cooperation on frontier AI risk assessments, accountability and incident reporting.
France launched the Paris Call for AI Safety and Security on 9 November, bringing together governments, AI companies, researchers and civil society organisations around a shared framework for addressing risks associated with advanced AI systems.
The declaration was signed by countries including France, Canada, Finland, the Netherlands and Kenya, alongside companies such as OpenAI, Anthropic, Microsoft, Mistral AI, Cohere and Hugging Face. It emerged from a meeting convened by French President Emmanuel Macron in Paris, attended by government representatives, international organisations, AI developers, researchers and civil society participants.
The initiative builds on previous international efforts, including the Hiroshima Process, the Seoul AI Summit’s Frontier Safety Commitments, discussions at the G7 meeting in Évian in June, and a September UN Security Council discussion on AI.
The declaration identifies three priorities for international cooperation: common approaches to evaluating and managing risks from frontier AI models, clearer accountability across the AI ecosystem, and stronger information-sharing and early-warning mechanisms. It argues that risks associated with advanced AI cross national borders and require cooperation between governments, companies and international organisations.
The first commitment focuses on developing a common framework for evaluating frontier AI risks, supported by compatible technical standards, testing tools and methodologies. This includes assessing potential misuse in areas such as cyberattacks and chemical, biological, radiological and nuclear threats, as well as risks linked to increasing model autonomy and potential loss of control. Signatories also support publishing non-sensitive evaluation findings in standardised formats and establishing a shared repository for results.
The second pillar addresses accountability. Developers are encouraged to disclose model capabilities, limitations and known risks, facilitate qualified independent evaluations where appropriate, and maintain mechanisms to restrict, pause or shut down systems under their control. The declaration also calls for continuous risk management, monitoring of emerging capabilities and clearer governance responsibilities.
The third area concerns information-sharing and incident response. Signatories support rapid alert systems and stronger reporting arrangements for serious AI and cyber incidents, including cooperation between governments and companies. The declaration also calls for secure channels through which whistleblowers can report concerns about systemic risks.
The launch coincided with broader security discussions in Paris. A separate session at the G7 Foreign Ministers’ Meeting addressed safe and trustworthy frontier AI, meaningful human involvement, malicious uses such as cybercrime, and international coordination. Japan also highlighted the Hiroshima AI Process and the growing security importance of undersea cables.
Why does it matter? The Paris Call seeks to move international AI safety discussions towards practical cooperation on model evaluations, accountability and incident reporting. However, it is a voluntary political commitment rather than a binding international regulatory framework. Its impact will depend on how signatories translate its principles into shared technical standards, reporting arrangements and national policies.
