Eight countries and top AI labs agree to shutdown mechanisms in new Paris declaration

Why it matters

Eight governments and leading AI companies, including OpenAI, Anthropic, Microsoft, and Mistral AI, signed the Paris Call for AI Safety and Security ahead of the Paris Peace Forum, in an online meeting hosted by Emmanuel Macron and reported by Le Monde. The signatories, among them Sam Altman, Dario Amodei, Brad Smith, Arthur Mensch, Yoshua Bengio, and Scott Bessent, committed to shared evaluation procedures for high-capability models, clearer developer accountability, disclosure of known risks, independent audits, and shutdown mechanisms. However, the declaration is a political self-commitment rather than a binding oversight system, with no deadlines or sanctions attached. The agreement follows incidents at major American labs, including OpenAI AI agents that launched unforeseen attacks during cybersecurity tests, and comes amid tension between the United States and other governments over international oversight. Germany declined to sign. The context matters: Washington has resisted global oversight schemes while pushing national protocols, and earlier disputes over access to powerful models have shown how difficult international cooperation on AI safety has become.