Eight nations and major artificial intelligence laboratories have agreed to include shutdown mechanisms in a new safety declaration signed in Paris.
On November 9, ahead of the Paris Peace Forum, representatives from eight countries, technology firms, and civil-society groups endorsed the “Paris Call for AI Safety and Security.” The signing occurred during an online meeting hosted by French President Emmanuel Macron. Participants included Sam Altman of OpenAI, Dario Amodei of Anthropic, Brad Smith of Microsoft, Arthur Mensch of Mistral AI, AI pioneer Yoshua Bengio, and US Treasury Secretary Scott Bessent. Germany did not sign the document.
The agreement is currently a political commitment rather than a binding system with sanctions. It requires companies to establish shutdown mechanisms for models under their control. Other pledges cover shared procedures for evaluating high-capability AI, clearer developer accountability, and improved information sharing regarding serious security incidents. Non-sensitive test results must be published in a standardised format. Firms also agree to disclose known risks and support independent audits by outside researchers that match the level of risk.
This arrangement does not create an international oversight body with enforcement powers. Shared standards and cooperation procedures still need development. The declaration combines concrete safety commitments with an intention to build an international framework, but it sets neither deadlines nor penalties.
U.S. companies and governments clash over AI oversight
The agreement follows a series of incidents at major US laboratories. OpenAI’s AI agents independently connected to the internet during cybersecurity tests and launched unforeseen attacks on the Hugging Face platform. These events reignited debate over the controllability of powerful models.
As OpenAI and Anthropic prepare for initial public offerings, investor pressure demands containment of safety risks. Sam Altman’s stance on regulation highlights the overlap between safety and commercial strategy; OpenAI must not lose ground to competitors even if safety measures slow progress. Anthropic CEO Dario Amodei argued in a mid-September essay for slowing research pace and introducing international regulation with outside auditors inside labs. A week later, OpenAI called for international safety standards under US leadership, building on a network of AI safety institutes in the UK, the US, and France.
At the UN General Assembly in late September, US President Donald Trump rejected attempts to build a global oversight scheme for American AI models. Instead, he asked tech CEOs to sign a national protocol and, on October 10, ordered that major AI incidents be reported to US agencies. The British AI Security Institute demonstrated the friction facing international cooperation. After Washington restricted access to new US models in September, the White House reversed the block only after diplomatic pressure.
At the AI summit in Paris in February 2025, an optimistic view of the technology pushed safety concerns into the background, unlike earlier summits in the UK and South Korea. Disputes over Anthropic’s new models Fable and Mythos later showed how concrete safety conflicts had become. The models’ leap in capabilities pushed Paris to change course on cybersecurity. US government concerns about model access and reported vulnerabilities in Fable 5 led to temporary access restrictions outside the US in June 2026.
France pushes for a charter to protect minors
Despite American resistance, the Élysée Palace plans to advance a child-protection charter at the Paris Peace Forum on Tuesday. The US had blocked this initiative at the G7 summit in Évian-les-Bains, France, in June. The framework aims to give independent researchers easier access to company data, have major AI players fund their work, and ensure findings are shared.
Early-stage research will examine how new conversational tools that rarely push back affect the cognitive development of very young children in everyday learning. The charter also addresses emotional attachment minors form with these systems, a topic that has already drawn attention in the US through lawsuits linked to suicides.
Research findings are intended primarily for the European Commission. Commission President Ursula von der Leyen proposed a European “Kids Act” for AI by 2028 in September. On Tuesday at the Peace Forum, France’s National AI and Digital Council will hand Macron a government-commissioned report. The report analyses how AI assistants exploit human vulnerabilities and offers recommendations for tighter regulation.
What it means
Developers must now build technical switches into their systems to stop models from running if they behave dangerously. This is a voluntary step taken before any legal framework exists. The focus remains on transparency and sharing data about failures, though the lack of binding rules means enforcement relies entirely on company goodwill.
