What happened
On 9 November 2026, Canada, France, Bahrain, Finland, Norway and the Netherlands signed a joint declaration on a “Paris Call for AI safety and security”, alongside ten AI companies and several civil-society organisations. The Prime Minister’s Office of Canada published the text the same day, and the declaration is non-binding. According to The Next Web, it was agreed at a meeting hosted by French President Emmanuel Macron on the opening day of the Paris Peace Forum.
The statement says it remains open to further endorsements and is being updated, so the list of signatories may change.
Key details
Who signed
- Governments: Bahrain, Canada, Finland, France, Norway, the Netherlands.
- Companies: Anthropic, Black Forest Labs, Cohere, Domyn, Hugging Face, Microsoft, Mistral AI, OpenAI, Sakana AI, Synthesia.
- Civil society: LawZero (Yoshua Bengio), SaferAI, AI Safety Connect and Ifri are listed in the Canadian text. The Next Web also reports that the French AI safety group CeSIA signed.
The United States is not a signatory, according to The Next Web, and neither are Google, Meta or China.
What the signatories commit to
The declaration frames three areas of consensus: a clear and predictable regulatory framework; prompt, tangible results on transparency, incident reporting and risk management; and rigorous evaluation standards. It then sets out three strategic commitments.
- A common evaluation and risk-management framework. Policymakers, researchers and developers commit to a shared framework built on existing ones, with common technical standards and tools. It focuses on misuse (including cyber and CBRN risks) and on risks from growing autonomy or misalignment, including loss of control. The text commits signatories to strengthen technical cooperation, including through the International Network for Advanced AI Measurement, Evaluation and Science (NAAIMES), to publish non-sensitive evaluation findings in a standardised format, and to create a publicly accessible shared repository for items such as red-teaming protocols, benchmarks and stress-test results.
- Accountability across the AI ecosystem. Actors are accountable according to their role. All commit to compliance with safety standards set or recognised by competent national authorities, with independent third-party evaluation. Policymakers are asked to clarify liability for large-scale harm, including cross-border harm. Frontier developers commit to disclose capabilities, limitations and known risks; support evaluation by qualified independent evaluators where appropriate; implement fail-safe shutdown mechanisms, including during training and internal use; address the risk of removed or circumvented safeguards while preserving the benefits of open-weight development; and maintain continuous risk management, monitoring, internal red-teaming, reporting lines that protect internal researchers, and independent oversight.
- Information sharing, reporting and early warning. Rapid alert systems are to build on existing incident-response arrangements, including cybersecurity ones. Signatories will align incident classification, share timely information on serious cyber and AI control incidents, and explore whether a cooperation mechanism would help. Policymakers commit to secure, confidential channels for whistleblowers raising systemic-risk concerns.
The declaration says it builds on the Hiroshima process, the Seoul Frontier Safety Commitments, the G7 Evian AI meeting, a UN Security Council meeting on AI on 23 September, and the “Call for Control of Frontier AI Models”. The Canadian page specifies no follow-up or review dates.
Criticism
BetaKit and Global News report that critics say the communiqué lacks teeth. CeSIA, which signed, says the declaration “leaves important gaps”, according to The Next Web: it sets no common risk thresholds for deciding whether a model should be developed or released, does not guarantee testers or governments access to the systems they need, and sets no deadlines. The shared testing framework it proposes has yet to be developed. CeSIA executive director Charbel-Raphaël Ségerie is quoted as saying: “These decisions must rest with public authorities.”
What it means for researchers and research offices
The declaration is a political statement, not a funding rule or regulation, and it creates no direct obligation for universities. The points below are our interpretation of where it could touch institutional practice.
- Evaluation research. The text names researchers as parties to the shared framework and NAAIMES as a vehicle for making evaluation methods comparable. Groups that build benchmarks or run red-teaming studies may find that standardised reporting formats and a shared repository change what counts as reusable evaluation output. Neither exists yet.
- Access and agreements. Commitments to support evaluation by qualified independent evaluators could affect negotiations over model access. Research offices reviewing agreements with AI developers may want to track how “qualified” is defined once frameworks emerge.
- Research security. Misuse risks named in the text (cyber and CBRN) and incident-sharing commitments sit close to the concerns of research-security programmes. Offices that handle dual-use review or security incidents involving AI systems may see incident classification schemes drawn from this work.
- Researcher protections. Reporting lines that protect internal researchers and confidential whistleblower channels are aimed at developers and policymakers, but they bear on how institutions think about staff who raise AI-safety concerns.
- Open-weight research. The wording on circumvented safeguards while preserving open-weight benefits flags a tension that affects institutions releasing models.
What to watch
- Whether more governments or companies endorse the text, and whether the US, Google or Meta join.
- Whether the common evaluation framework and the shared repository are actually built, and by whom.
- Whether NAAIMES publishes technical standards or reporting formats.
- Whether national authorities turn these commitments into binding requirements, given that no deadlines are set.
Sources
- Prime Minister of Canada, “Joint declaration on Paris call for AI safety and security”, 9 November 2026 (read 13 November 2026)
- The Next Web, Paris Call on AI safety coverage (read 13 November 2026)
- BetaKit, “Government of Canada signs international declaration on AI safety and security” (read 13 November 2026)
