Guidelines  ·  2026-11-15

Paris Call for AI Safety and Security — joint declaration by six governments and ten frontier AI companies

GuidelinesHigh impactFrance
On 9 November 2026, at the Paris Peace Forum, six governments, ten frontier AI companies and several civil-society organisations signed the non-binding 'Paris Call for AI Safety and Security' joint declaration (text published by Canada's Prime Minister's Office the same day). It sets out three areas of consensus and three strategic commitments: (1) a common evaluation and risk-management framework built on existing ones, with common technical standards/tools focused on misuse (including cyber and CBRN risks) and on autonomy/misalignment/loss-of-control; (2) role-based accountability, with frontier developers committing to disclose capabilities/limitations/known risks, support independent third-party evaluation and implement fail-safe shutdown mechanisms, and policymakers committing to clarify liability for large-scale harm; and (3) information sharing, reporting and early warning — aligning incident classification and sharing timely information on serious cyber and AI control incidents, plus confidential whistleblower channels. It remains open to further endorsements and has no deadlines (critics, including signatory CeSIA, note no common risk thresholds, no guaranteed tester access, and no timeline).
This is the latest multilateral convergence on AI security evaluation and incident-reporting standards, explicitly naming cyber misuse and loss-of-control among the risks to be evaluated. It signals the direction of travel for shared technical standards, standardised evaluation-reporting formats, and a shared repository of red-teaming protocols, benchmarks and stress-test results — the very substance of AI-security practice. Even though non-binding, it binds the leading frontier labs and six states to a common framework that practitioners may be asked to align with, and it reinforces the NAAIMES evaluation network.
Monitor for the follow-on shared evaluation/risk-management framework and any standardised incident-classification or evaluation-reporting format that emerges; map the declaration's evaluation and incident-reporting commitments onto your AI governance and red-teaming programme so you are ready to align if endorsed by your jurisdiction or a customer.
Joint declaration on Paris call for AI safety and security (full text via CNW)Paris Call on AI Safety and Security Explained (CASRAI)
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →