Canada signs Paris call on AI safety with five countries and firms including OpenAI and Cohere

Canada has signed a new international declaration committing governments and major artificial intelligence companies to shared safety testing of the most powerful AI systems, along with shut-off mechanisms and early warnings when something goes seriously wrong.

The Paris Call for AI Safety and Security, dated Monday and published by the Prime Minister’s Office and the French presidency, was signed by the leaders of Bahrain, Canada, Finland, France, Norway and the Netherlands. Company signatories are Anthropic, Black Forest Labs, Cohere, Domyn, Hugging Face, Microsoft, Mistral AI, OpenAI, Sakana AI and Synthesia. Civil society signatories include LawZero, the group led by Canadian AI researcher Yoshua Bengio.

The list published Monday does not include the United States, the United Kingdom, China, Google or Meta. The statement says it remains open to new endorsements and is being updated.

What they promised

The declaration sets out three commitments. The first is a common framework for testing and managing the risks of frontier models, focused on dangerous misuse, including cyberattacks and chemical, biological, radiological and nuclear threats, and on risks from growing autonomy or loss of control. Signatories agree to publish the non-sensitive findings of their evaluations in a standard format and to create a shared public repository.

The second covers accountability. Developers commit to disclose what their models can do, their limits and known risks, and to build mechanisms to shut down operations within their control, including during training and internal use. Governments recognize the need for laws that clarify who is responsible when AI causes large-scale damage, including across borders.

The third is about warning systems. Signatories back rapid alert systems and better sharing of information on serious cyber and AI control incidents. Governments commit to offer secure, confidential channels for whistleblowers to raise concerns about systemic risks without retaliation.

What it does not say

The text names no enforcement body, penalties or deadlines. Several of its provisions are tentative: independent evaluation is to be supported “where appropriate and proportionate to the risk,” and signatories say only that they will explore whether a cooperation mechanism on incident reporting would help.

For Canada, the signature puts the federal government behind a commitment to legal clarity on AI harms and whistleblower protection. The statement released by the Prime Minister’s Office does not say how or when Ottawa intends to act on either.

Source: Élysée and the Prime Minister’s Office.

Image: Présidence de la République.