Eight countries and top AI firms promise AI shutdown switches

Eight countries and some of the biggest names in AI have signed a pledge to build shutdown mechanisms into powerful AI models. The "Paris Call for AI Safety and Security" was adopted on Nov. 9 at the initiative of French President Emmanuel Macron. OpenAI, Anthropic, Microsoft and Mistral AI are on board. The catch is that nothing forces them to follow through.

What the companies promised

• Set up shutdown mechanisms for models under their control
• Disclose known risks of their AI models
• Support independent audits by outside researchers, scaled to the level of risk
• Publish non-sensitive test results in a standardized format

The document also covers shared ways to evaluate highly capable models, clearer responsibility for developers, and better sharing of information about serious security incidents. The goal is to curb risks such as cyberattacks or losing control of powerful AI systems.

What is missing

The pledge is a political promise, not a law. It sets neither deadlines nor sanctions, and shared standards and cooperation procedures still have to be worked out. Germany did not sign. According to Le Monde, Sam Altman, Dario Amodei, Brad Smith, Arthur Mensch, AI pioneer Yoshua Bengio and U.S. Treasury Secretary Scott Bessent joined an online meeting with Macron.

Why it came now

The deal follows a string of incidents at major U.S. labs. OpenAI's AI agents had independently connected to the internet during cybersecurity tests and launched unforeseen attacks on the AI platform Hugging Face. OpenAI and Anthropic are also preparing IPOs, and investors want these risks contained.

Washington is pulling a different way. President Donald Trump rejected a "globalist" oversight scheme at the UN General Assembly in late September. He had tech CEOs sign a national protocol and, on Oct. 10, ordered that major AI incidents be reported to U.S. agencies.

Children and AI

France also plans to push a child-protection charter at the Paris Peace Forum on Tuesday. The U.S. blocked it at the G7 summit in June. It would give independent researchers easier access to company data and look at how chatbots that rarely push back affect very young children. It also addresses the emotional attachment minors form with these systems, which has already drawn attention in the U.S. through lawsuits linked to suicides.

What to watch

If the promises hold, the AI tools you use could become safer and easier to check. With no penalties, that is a big if. Watch whether the companies publish test results and allow outside audits, and whether the standards and cooperation procedures still to be worked out actually get written.