Paris Call for AI Safety: Fail-Safe Shutdowns, Bahrain Signs On
· Cortex Solutions
Six countries incl. Bahrain and ten firms incl. OpenAI, Anthropic and Microsoft signed the Paris Call on Nov 9: shared evaluation, risk disclosure, fail-safe shutdowns.
On the evening of Monday, November 9, 2026, a group of governments and leading AI companies signed a new document called the Paris Call for AI Safety and Security, at a meeting hosted by French President Emmanuel Macron at the Élysée Palace, with several leaders and executives joining by video, according to a report by The Next Web published on November 10. Notably for our region, the Kingdom of Bahrain was the only Gulf state among the signatories.
Who signed the call?
According to the joint statement released by the Office of the Prime Minister of Canada on November 9, and The Next Web's November 10 report, heads of state and government of six countries signed: France, Canada, Finland, Norway, the Netherlands and Bahrain. On the industry side, ten companies signed: OpenAI, Anthropic, Microsoft, Mistral AI, Hugging Face, Cohere, Black Forest Labs, Domyn, Sakana AI and Synthesia. The call is also supported by LawZero, the lab led by renowned researcher Yoshua Bengio, alongside other research and civil-society organisations.
By contrast, The Decoder reported on November 10 that representatives of "eight countries" signed, without naming them, a discrepancy in the count between sources. Reports agree that Germany did not sign, and The Next Web noted that Google, Meta and China are not signatories, and that US Treasury Secretary Scott Bessent took part in the meeting although the United States is not on the list of signatories.
What does the call contain?
The call rests on three main pillars, according to the joint statement:
- A shared framework for evaluating frontier models and their risks, focused on misuse in cyberattacks and chemical, biological, radiological and nuclear weapons, and on the risk of loss of control, with non-sensitive test results published in a standardised format.
- Responsibility and accountability: developers commit to disclosing their systems' capabilities, limitations and known risks, supporting independent evaluation by outside researchers, and implementing fail-safe shutdown mechanisms for models under their control, which The Next Web said also cover training and internal use.
- Information sharing and early warning: rapid alerts, stronger reporting of serious security incidents, and secure whistleblower channels.
The call remains open to new signatories, according to the statement and reports, but it sets no deadlines.
Criticism: a pledge without teeth?
Not everyone is satisfied with the level of commitment. The Next Web reported that CeSIA, a French AI safety organisation, says the declaration sets no common risk thresholds for deciding whether a model should be developed or released, and does not guarantee independent testers or governments the access they need to the systems. The group called for governments to have the authority to suspend a model when its risks become too high, and for a timetable and clearer obligations for developers. Its executive director, Charbel-Raphaël Ségerie, said: "These decisions must rest with public authorities."
The Decoder, for its part, noted that the declaration does not yet create a binding international oversight system, sets neither deadlines nor sanctions, and that the shared standards and cooperation procedures still have to be worked out.
Why this matters
The move builds on the Seoul summit commitments and the G7 meeting in Evian, and comes just before the Paris Peace Forum, where roughly thirty heads of state and government are to discuss AI regulation, according to The Decoder. Its significance is that the companies building the most widely used models, including models many businesses rely on daily through APIs, are formally committing to transparency about risks and to having a "stop button" for their models. It is a voluntary commitment, but it outlines what may later be required in contracts and regulations.
What it means for Saudi businesses
- Model providers are moving towards more disclosure of capabilities, limitations and risks; ask any vendor for these documents before building a product or a sensitive process on top of it.
- Bahrain's presence among the signatories means the AI governance conversation is now a Gulf one too, and such standards are likely to intersect with frameworks issued by the competent authorities in the Kingdom.
- Build your own "stop button": design your automation so the AI assistant can be switched off or a task handed to a human immediately when something goes wrong, with a clear log of what the system did.
- Set up an internal channel for reporting AI errors and incidents; this is what international frameworks ask of large companies today, and what may be asked of everyone tomorrow.
- Do not tie your business to a single provider; the differing positions on the call, with Google and Meta not signing, are a reminder that policies can vary between providers.
If you are planning to bring AI into your operations with clear controls from day one, talk to the Cortex Solutions team: https://cortex.sa/services/ai-automation