Paris Call: 17 countries, 10 firms pledge AI off-switches
Anthropic, OpenAI and Microsoft have promised to be able to shut down their models, with no deadlines or penalties. Mexico did not sign.
Companies that build frontier AI models will need to be able to switch them off. Nobody will make them do it. That is the core of the Paris Call for AI Safety and Security, signed by the French presidency on Nov. 9 and now endorsed by 17 governments and 10 companies.
The companies include Anthropic, OpenAI and Microsoft, along with Cohere, Mistral AI, Hugging Face, Sakana AI, Black Forest Labs, Domyn and Synthesia. The governments are Austria, Bahrain, Canada, Croatia, Cyprus, Estonia, Finland, France, Germany, Ireland, Kenya, Latvia, Liechtenstein, Luxembourg, the Netherlands, Norway and Portugal. The list is still open and has already grown: early reports counted 16 governments.
What they promise
The central pledge is "fail-safe mechanisms" able to shut down a model for as long as it stays "within their control." The text names restrictions, pauses and shutdowns, in training as well as internal use. Signatories add three more commitments:
- Share timely information on serious cyber and AI control incidents.
- Offer confidential channels for people who raise alarms about systemic risks, without legal or professional retaliation.
- On the governments' side, clarify who is responsible for large-scale damage, including harm that crosses borders.
What it leaves out
The document sets no deadlines and no penalties, so everything rests on each company's goodwill. And "within their control" excludes whatever developers no longer control: copies of open models that third parties download and run on their own.
That makes the practical reach modest. No outside authority can order a model shut down on the strength of this text, and nobody checks that the mechanism exists.
For Mexico
Nothing changes right away. Mexico is not among the signatory governments, and the text creates no obligations for users or companies there. Google, Meta and the Chinese AI companies are not on the list either.
What can still move is that list: the Élysée says the text remains open to new endorsements and is being updated.
Keep reading
- Artificial intelligence
Four nonprofits sue to halt the most advanced AI
- Artificial intelligence
OpenAI says Control Room halts research AI jobs in 35 seconds
- Internet
Microsoft probes Exchange Online outage hitting Outlook users
- Artificial intelligence
OpenAI's ChatGPT Wallet lets AI agents pay with a card



