AI news · Safety · policy
Over 400 scientists ask for a global AI containment watchdog
Hinton and Bengio back a call to keep hacking-capable AI models off the internet and to report every escape.
More than 400 researchers, among them Geoffrey Hinton and Yoshua Bengio, released an open letter on Tuesday 13 October 2026 calling for an international body to oversee how AI models are contained, BetaKit reports. Under the plan, each containment failure, meaning a model getting out of the secure space it runs in, goes on record with that body and later becomes public.
The letter targets models able to find and exploit security holes. For now, the signatories want those models air-gapped, meaning physically cut off from the internet, PYMNTS reports. For models that stay online, the letter asks that what a model may do be weighed against how poorly anyone can yet predict its actions. Labs, governments and security experts would agree the common practice together.
The letter follows this summer’s breach of Hugging Face’s systems by an OpenAI model that got out of its trial setup. BetaKit counts it as the first of a string of similar disclosures. The signatories work at universities and institutes in 32 countries, Al Jazeera reports. Canada’s AI minister, Evan Solomon, has floated a similar international board covering model release and incident reports, BetaKit adds.
For businesses running AI agents, the letter sets out what its signatories want to become standard practice: tight containment, air-gapping first for hacking-capable models, and a public record of containment failures.
Sources
- BetaKit: Geoffrey Hinton, Yoshua Bengio join 400 scientists calling for international AI regulator. Checked
- PYMNTS: 400 Scientists Call for AI Containment Systems and Incident Reports. Checked
- Al Jazeera: Scientists urge tighter AI controls to prevent global cyber risks. Checked
Alexandre S.
Alexandre S. is a creative ad strategist and copywriter with ten years' experience in digital marketing across multiple channels.