Ten AI companies, including OpenAI, Anthropic, Microsoft and Mistral, have signed the Paris Call for AI Safety and Security at a Paris summit hosted by President Emmanuel Macron. The declaration, also signed by six countries, sets out commitments to improve independent risk evaluation, share incident alerts, and allow expert testing of advanced AI models. Signatories have agreed to develop fail-safe mechanisms to shut down models and publish non-sensitive test results, but major players like Google, Meta and China have not joined.
The agreement follows earlier safety initiatives and is open to further signatories, but sets no deadlines or binding risk thresholds. French AI safety group CeSIA, which signed the declaration, criticised the lack of clear obligations, enforceable standards and government authority to suspend risky models. CeSIA called for public authorities to set the rules and for a timetable to be established.
For regulated firms, the declaration signals growing pressure for independent oversight and transparency in AI deployment. The absence of binding rules and the limited group of signatories mean regulatory expectations may shift further as governments and industry debate enforcement and access. The framework for shared testing and incident reporting is still to be developed.
Macron also called for pre-certification of AI tools used by children, likening it to toy safety checks, and warned of the mental health risks posed by unregulated AI systems for young users. A charter on safer AI for children was expected to follow.
