Heads of state and government from six countries signed a joint declaration on AI safety and security under the Paris call on Monday, November 9, 2026, according to a release issued by the Prime Minister's Office of Canada. The signatory governments listed at publication are Bahrain, Canada, Finland, France, Norway and the Netherlands.
Ten AI companies also endorsed the text: Anthropic, Black Forest Labs, Cohere, Domyn, Hugging Face, Microsoft, Mistral AI, OpenAI, Sakana AI and Synthesia. Civil society signatories include LawZero, the organization led by Yoshua Bengio, along with SaferAI, AI Safety Connect and the French think tank Ifri. The release says the declaration remains open to further endorsements.
The first commitment is to build a common framework for evaluating and managing risks from frontier models, drawing on existing frameworks and focusing on misuse such as cyber and CBRN threats as well as autonomy and loss-of-control risks. Signatories say they will publish non-sensitive evaluation findings in a standardized format and create a publicly accessible shared repository. The text points to the International Network for Advanced AI Measurement, Evaluation and Science (NAAIMES) as a way to make evaluation methods and results more comparable.
On accountability, governments commit to clarifying legal responsibility for large-scale damages, including cross-border harm. Frontier developers commit to disclosing capabilities, limitations and known risks, supporting independent evaluation where proportionate, implementing fail-safe shutdown mechanisms, and addressing the risk of safeguards being stripped from models, including through "abliteration", while preserving the benefits of open weights.
The declaration also calls for rapid alert systems built on existing incident-response and cybersecurity arrangements, aligned classification of serious cyber and AI control incidents, and secure whistleblower channels provided by policymakers. It cites the Hiroshima process, the Seoul Frontier Safety Commitments and the G7 Evian AI meeting as foundations.
The commitments are framed as intentions rather than binding obligations, and the release names no enforcement mechanism. For developers, the most concrete near-term items are the planned shared evaluation repository and a standardized format for publishing evaluation results, which could shape how model cards and safety reports are written.