Six governments, ten AI developers and four safety groups signed a joint declaration in Paris on 9 November, committing to shared ways of testing the most powerful AI models and warning each other about serious incidents. The governments are Bahrain, Canada, Finland, France, Norway and the Netherlands. The developers include Anthropic, OpenAI, Microsoft, Mistral AI, Cohere and Hugging Face, and the civil society signatories include Yoshua Bengio’s LawZero.
Signatories commit to build a common framework for evaluating frontier models, focused on misuse for cyber attacks and chemical, biological, radiological and nuclear weapons, and on loss of control. Non-sensitive results are to be published in a standard format in a public repository. Developers commit to disclose capabilities and known risks, support independent evaluation and build mechanisms to shut down model operations within their control. Governments commit to clarify liability for large-scale harm, including across borders, and to offer confidential channels for whistleblowers. All sides will work on rapid alert systems for serious incidents.
The declaration, released by Canada’s Prime Minister’s Office, is voluntary and remains open to new signatories. Google, Meta and xAI are not on the list, and nor are the United States or China.
Why it matters: it is a voluntary pledge, but it puts specific promises on testing, shutdown and incident sharing on the record.