Eight countries and representatives from OpenAI, Anthropic, Microsoft and Mistral AI have signed the Paris Call for AI Safety and Security, a voluntary declaration aimed at high-capability systems.
The document calls for shared model-evaluation procedures, clearer developer accountability and better exchange of information about serious security incidents. Signatories also commit to publish non-sensitive test results in a standardized form, disclose known risks, support independent audits proportionate to risk and maintain shutdown mechanisms for models under their control.
The agreement was adopted on November 9 ahead of the Paris Peace Forum. Participants in the online meeting with French President Emmanuel Macron included the companies’ leaders, AI researcher Yoshua Bengio and U.S. Treasury Secretary Scott Bessent. Germany did not sign, according to the report.
For now, the declaration is a political commitment rather than an enforcement system. Common standards and cooperation procedures still need to be developed, and the text sets neither deadlines nor sanctions. Its significance will therefore depend on whether governments and labs turn the stated controls into auditable practices and whether outside researchers receive enough access to test them.