Ten AI Companies Including OpenAI, Anthropic and Mistral Sign the Élysée’s Paris Call for AI Safety and Security
Ten artificial intelligence companies, including OpenAI, Anthropic and France’s Mistral AI, have signed a text drawn up by the Élysée on the security risks of the most advanced AI models, which the presidency described as a first. The Paris Call for AI Safety and Security was signed on Monday, November 9, after Emmanuel Macron convened a meeting with heads of state and government, AI company leaders including Sam Altman, Dario Amodei and Arthur Mensch, and researchers such as Yoshua Bengio. Bahrain, Canada, Finland, France, Norway and the Netherlands signed, as did Black Forest Labs, Cohere, Domyn, Hugging Face, Microsoft, Sakana AI and Synthesia.
According to the Élysée communiqué, the talks produced three points of consensus: the need for independent evaluators, alignment on standards for evaluating models and reporting major incidents, and clearer responsibility for each actor. Signatories commit to a common framework for assessing frontier-model risks, built on existing frameworks and on the International Network for Advanced AI Measurement, Evaluation and Science (NAAIMES), with particular attention to cyber and chemical, biological, radiological and nuclear misuse and to loss of control. Non-sensitive evaluation results are to be published in a standard format in a shared public repository.
What it means in practice
Developers pledge to disclose the capabilities, limits and known risks of their models, support evaluation by qualified independent evaluators, build mechanisms that can stop a model from running and strengthen internal governance. Governments promise to clarify liability for large-scale damage, including across borders, and to offer secure, confidential channels for whistleblowers. Henry Papadatos, executive director of the French association SaferAI, told franceinfo it was an important moment for AI safety and for French leadership, and urged a coalition of countries to turn the priorities into concrete measures within months.
The call builds on work at home such as Inria’s €13.5 million AI evaluation and safety programme and came a day before an expert commission handed Macron 18 recommendations on generative AI. The text does not spell out how the commitments would be enforced. What to watch is whether more states and companies sign, since the declaration remains open, how the shared repository of evaluation results is set up, and whether a dedicated cooperation mechanism for incident alerts emerges.