OpenAI, Anthropic, Microsoft sign Paris AI safety pact

Six governments and ten AI companies, including Anthropic, OpenAI and Microsoft, signed the "Paris Call for AI Safety and Security" on Monday, November 9. The declaration commits them to shared evaluations of frontier models, incident early-warning systems and clearer accountability for developers, according to the Élysée Palace. The national signatories are Bahrain, Canada, Finland, France, Norway and the Netherlands. The other company signatories are Black Forest Labs, Cohere, Domyn, Hugging Face, Mistral AI, Sakana AI and Synthesia. The declaration is not binding and remains open to new endorsements, the text states.elysee.fr+1

A gathering called by Macron

French President Emmanuel Macron brought together about 50 participants by video link, among them world leaders, researchers and heads of AI labs, dpa reported. Leaders from the United Kingdom, Japan and Sierra Leone took part in the talks but were not listed as signatories. U.S. Treasury Secretary Scott Bessent and representatives of the United Nations and the OECD also took part, as did Sam Altman of OpenAI, Dario Amodei of Anthropic and Microsoft President Brad Smith, Logistik Heute reported, citing dpa. Macron said the participants had pledged to assess the risks of the most advanced models independently and publicly, share warnings about serious incidents more effectively and hold developers accountable, according to dpa. Civil society groups also signed, including Yoshua Bengio's LawZero, SaferAI, AI Safety Connect and the French think tank Ifri, the Élysée said.dpa-international.com+2

Three commitments

The first commitment is a common framework for evaluating and managing the risks of frontier models. It focuses on dangerous misuse in the cyber and CBRN (chemical, biological, radiological and nuclear) domains, and on risks from growing autonomy, misalignment and loss of control. Signatories agreed to publish the non-sensitive findings of their evaluations in a standardized format in a shared public repository, according to the declaration.presidentti.fi

The second covers responsibility. Developers commit to disclosing model capabilities, limitations and known risks, and to supporting evaluation by independent evaluators where appropriate. They also commit to building fail-safe mechanisms that can pause or shut down models within their control, including during training and internal use. Policymakers, for their part, pledged to clarify who is liable when AI causes large-scale damage, including harm that crosses borders, the text says.presidentti.fi

The third calls for rapid alert systems and better sharing of information on serious cyber and AI control incidents. It also commits governments to offer secure channels for whistleblowers, according to the Élysée.elysee.fr

A tense security backdrop

The declaration came a week after a suspected cyberattack on November 2 knocked out supermarket refrigeration systems in 11 European countries, Türkiye Today reported. The same report noted that the cybersecurity firm Wiz called October "Black October" after at least nine AI development platforms disclosed security incidents. It added that authorities have not established whether the attacks were linked or whether AI played a role in every case.turkiyetoday.com

The Élysée said the charter is meant to inform national and European legislative debates, Pollar reported.pollar.news