The Pulse
Governments and AI companies sign Paris safety declaration
A declaration signed by governments, AI companies and civil society groups sets out three commitments on frontier AI safety. Shared evaluation methods, stronger accountability and better systems for reporting incidents and emerging risks ar
AI.info Team · Written with AI · fact-checked ·
A joint declaration published by the Prime Minister of Canada’s office sets out commitments on frontier AI safety, accountability and information sharing. It was signed on Monday, November 9, 2026, by listed government leaders, AI companies and civil society representatives, and remains open to further endorsements.
The governments listed as signatories are Bahrain, Canada, Finland, France, Norway and the Netherlands. The company signatories named are Anthropic, Black Forest Labs, Cohere, Domyn, Hugging Face, Microsoft, Mistral AI, OpenAI, Sakana AI and Synthesia. Civil society signatories include LawZero, represented by Yoshua Bengio; SaferAI, represented by Henry Papadatos; AI Safety Connect, represented by Nicolas Miailhe; and Ifri, represented by Laure de Roucy-Rochegonde.
Shared evaluation standards
The declaration reflects, the signatories say, a shared interest in developing AI that is safe, secure, useful and widely available. Frontier models have global development, deployment and impacts, the declaration says, and no single state, company or international organization can manage all the risks alone.
The three priorities identified from discussions are a clear and predictable regulatory framework, immediate progress on transparency and incident reporting, and rigorous evaluation standards. The first commitment is for policymakers, researchers and developers to establish a common framework for evaluating and managing risks from frontier models, building on existing frameworks.
The proposed framework would draw on common technical standards, tools and methods that can keep pace with changes in model capabilities. Risks involving dangerous misuse, including in cyber and CBRN domains, as well as growing autonomy, model misalignment and loss of control, are highlighted in the declaration. The signatories agree to publish non-sensitive evaluation findings in a standardized format and create a publicly accessible repository that the declaration says could include red-teaming protocols, benchmarking tools, risk assessments and stress-test results.
Responsibilities across the model lifecycle
Responsibility and accountability should apply to all actors in the AI ecosystem according to their roles, the second commitment affirms. Across the model lifecycle, the signatories commit to comply with rigorous safety standards set or recognized by national authorities, with oversight by those authorities and evaluation by credible, competent and independent third parties.
For policymakers, legal frameworks that clarify responsibility and accountability in cases of large-scale harm, including when it crosses borders, are called for, with the declaration saying this should protect affected people and provide AI companies with legal certainty for economic development.
Developers of frontier models commit to disclose capabilities, limitations and known risks, and, when appropriate and proportionate to the risk, to support evaluation by qualified independent evaluators. They also commit to fail-safe mechanisms that can stop model operations under their control, including during training and internal use. The risk of safeguards being removed or circumvented, including through abliteration, is noted in the declaration, while it also recognizes the benefits of open-weight development for research, innovation and safety.
Internal governance, the declaration calls for, must require continuous risk management, real-time monitoring of emerging capabilities, internal red-teaming and reporting lines that protect researchers. Developers are committed, under the declaration, to working on governance structures with defined responsibilities, adequate resources and independent oversight.
Incident reporting and early warnings
The third commitment covers information sharing, reporting and early-warning systems. Rapid alerts, building on existing incident-response arrangements including cybersecurity arrangements, are essential to prevent cascading systemic risks and respond to emerging threats, the text says.
Strengthening reporting arrangements is also proposed so organizations can align incident classifications and share timely information on serious cyber and AI-control incidents and other signals that could precede serious harm. Information sharing could take place between companies and governments, between governments, or between AI companies, and the declaration calls for exploring whether a cooperation mechanism would help.
Policymakers commit to offering secure, confidential channels for whistleblowers, including employees and other relevant stakeholders, to raise concerns about systemic risks on reasonable grounds without fear of legal or professional retaliation.