Governments and AI Firms Sign Safety Declaration

Canada, Bahrain, Finland, France, Norway and the Netherlands joined AI companies and civil society groups in signing a declaration on Monday, November 9, 2026, calling for cooperation on AI safety and security. The statement remains open to new endorsements and is being updated. Signatory companies include Anthropic, Black Forest Labs, Cohere, Domyn, Hugging Face, Microsoft, Mistral AI, OpenAI, Sakana AI and Synthesia. Civil society signatories include LawZero, SaferAI, AI Safety Connect and Ifri.

The declaration says frontier models are developed, deployed and used across national borders, so their opportunities and risks are global. It calls for a clear and predictable regulatory framework, immediate progress on transparency, incident reporting and risk management, and rigorous evaluation standards. It cites earlier cooperation through the Hiroshima process, the Frontier Safety Commitments of the Seoul AI Summit, the G7 Evian AI meeting, a United Nations Security Council meeting on AI held on September 23, and the “Call for Control of Frontier AI Models.”

The signatories commit to a shared framework for evaluating and managing risks from frontier models, supported by common technical standards, tools and methods. Priorities include dangerous misuse in cyber and CBRN domains, growing autonomy and model misalignment, including loss of control. They also back publishing non-sensitive evaluation findings in a standard format and creating a public repository that could hold red-teaming protocols, benchmarking tools, risk assessments and stress-test results. The declaration notes that the International Network for Advanced AI Measurement, Evaluation and Science (NAAIMES) could help make evaluation methods and results more comparable and accessible across regions.

The declaration assigns responsibilities across the AI ecosystem. Policymakers are urged to clarify legal responsibility for large-scale harms, including cross-border damage. Developers of frontier models commit to disclose capabilities, limitations and known risks; support qualified independent evaluations where appropriate; and use fail-safe mechanisms to restrict, pause or shut down operations under their control, including during training and internal use. They also acknowledge risks from removing or circumventing safeguards, including “abliteration,” while supporting the benefits of open-weight development for research, innovation and safety. The text calls for ongoing risk management, real-time monitoring of emerging capabilities, internal red-teaming, protected reporting lines and independent oversight.

The third commitment is information sharing, reporting and early warning. The signatories call for rapid alert systems built on existing incident-response arrangements to address emerging threats and reduce their real-world effects. They also propose aligning incident classifications and sharing timely information about serious cyber and AI-control incidents and warning signs of serious harm among companies and governments. Policymakers commit to secure, confidential channels where employees and other stakeholders can report systemic risks without fear of legal or professional retaliation.

Canada, Bahrain, Finland, France, Norway and the Netherlands joined AI companies and civil society groups in signing a declaration on Monday, November 9, 2026, calling for cooperation on AI safety and security. The statement remains open to new endorsements and is being updated. Signatory companies include Anthropic, Black Forest Labs, Cohere, Domyn, Hugging Face, Microsoft, Mistral AI, OpenAI, Sakana AI and Synthesia. Civil society signatories include LawZero, SaferAI, AI Safety Connect and Ifri.

The declaration says frontier models are developed, deployed and used across national borders, so their opportunities and risks are global. It calls for a clear and predictable regulatory framework, immediate progress on transparency, incident reporting and risk management, and rigorous evaluation standards. It cites earlier cooperation through the Hiroshima process, the Frontier Safety Commitments of the Seoul AI Summit, the G7 Evian AI meeting, a United Nations Security Council meeting on AI held on September 23, and the “Call for Control of Frontier AI Models.”

The signatories commit to a shared framework for evaluating and managing risks from frontier models, supported by common technical standards, tools and methods. Priorities include dangerous misuse in cyber and CBRN domains, growing autonomy and model misalignment, including loss of control. They also back publishing non-sensitive evaluation findings in a standard format and creating a public repository that could hold red-teaming protocols, benchmarking tools, risk assessments and stress-test results. The declaration notes that the International Network for Advanced AI Measurement, Evaluation and Science (NAAIMES) could help make evaluation methods and results more comparable and accessible across regions.

The declaration assigns responsibilities across the AI ecosystem. Policymakers are urged to clarify legal responsibility for large-scale harms, including cross-border damage. Developers of frontier models commit to disclose capabilities, limitations and known risks; support qualified independent evaluations where appropriate; and use fail-safe mechanisms to restrict, pause or shut down operations under their control, including during training and internal use. They also acknowledge risks from removing or circumventing safeguards, including “abliteration,” while supporting the benefits of open-weight development for research, innovation and safety. The text calls for ongoing risk management, real-time monitoring of emerging capabilities, internal red-teaming, protected reporting lines and independent oversight.

The third commitment is information sharing, reporting and early warning. The signatories call for rapid alert systems built on existing incident-response arrangements to address emerging threats and reduce their real-world effects. They also propose aligning incident classifications and sharing timely information about serious cyber and AI-control incidents and warning signs of serious harm among companies and governments. Policymakers commit to secure, confidential channels where employees and other stakeholders can report systemic risks without fear of legal or professional retaliation.