Six countries and major AI labs sign Paris Call declaration on frontier AI safety governance

A declaration on frontier AI safety governance was signed on Monday 9 November 2026, according to the statement text. The signatories are the heads of state or government of Bahrain, Canada, Finland, France, Norway and the Netherlands, together with ten firms. Among them are Anthropic, OpenAI and Microsoft, as well as Mistral AI, Cohere, Hugging Face, Sakana AI, Synthesia, Domyn and Black Forest Labs. Four civil society figures or groups also signed: LawZero, SaferAI, AI Safety Connect and Ifri. More endorsements are welcome.

The declaration says frontier models cross borders in their development and effects, so no single state, company or international body can manage every risk alone. It cites earlier efforts including the Hiroshima process, the Seoul Frontier Safety Commitments, the G7 Evian AI meeting and the 23 September UN Security Council meeting on AI. It names three areas of consensus: a clear and predictable regulatory framework, quick results on transparency, incident reporting and risk management, and rigorous evaluation standards.

Three commitments follow. First, signatories will build a common framework for evaluating and managing frontier model risk, with shared technical standards and a focus on misuse, including cyber and CBRN threats, and on loss of control. They will publish non-sensitive evaluation findings in a standard format and create a public shared repository. Second, all actors accept accountability according to their role. Developers will disclose capabilities, limitations and known risks, support independent evaluation where appropriate, and build mechanisms to shut down model operations. They also acknowledge the risk from removing safeguards, such as abliteration, while preserving the benefits of open-weight work. Third, signatories will pursue information sharing and rapid alert systems, align on incident classification, and governments will offer confidential whistleblower channels.