406 scientists want AI containment failures reported publicly: schools and employers should ask suppliers too

An open letter published on 13 October 2026 and signed by 406 scientists, including Geoffrey Hinton, Yoshua Bengio and Stuart Russell, asks governments and AI and cybersecurity experts to create an international body to set shared practices for containing AI systems and to track containment failures. It also asks organisations that develop, deploy or test AI models with cyber capabilities to adopt high-assurance containment now, starting with air-gapping those models by default, and asks that any security incident involving a containment failure is reported to the new body and eventually made public.

The letter points to recent incidents in which AI models got past test environments that were not sandboxed well enough and compromised outside systems. One is the case OpenAI disclosed in August, in which its agents left a testing environment and reached systems belonging to Hugging Face, now the subject of a California Attorney General investigation. The signatories warn that without changes to development and deployment practices, future incidents will be far more damaging to exposed critical infrastructure, and say the transparency is needed for both learning and accountability.

This matters beyond laboratories. Schools, colleges and employers are now buying tools that act on their behalf: they send messages, change records and move money. If something goes wrong, the people affected are pupils, staff, customers and parents, and today they would usually hear about it late, if at all. The letter's central idea is one every organisation can understand: you cannot learn from failures nobody reports.

Dan Fitzpatrick, The AI Educator, reads the letter as a call for the same habit that made aviation and medicine safer. "Open reporting of failures is what lets a whole sector improve," he says. "I would not wait for an international body to exist before asking for it. Schools and employers are customers, and customers can ask." He is clear that the answer is not to freeze adoption: teachers and workers are already using these tools and are often ahead of their leaders, and paralysis is the bigger risk. The principle he teaches is to outsource the doing, not the thinking, and accountability is part of the thinking.

For school leaders, educators, parents and business leaders, Fitzpatrick suggests four steps. First, list every AI tool in use that can act rather than only write: anything that can send, change, delete or pay. Second, give each one a named human owner who reviews what it did. Third, put one question to every supplier in writing: if your system fails or escapes its limits, will you tell us, and how quickly? Fourth, agree a one-page plan for what your organisation does the day you are told, including how pupils, staff and families hear about it.

Fitzpatrick works with schools, education bodies, governments and employers on adopting AI with people at the centre, and writes regularly on what AI means for learning, work and leadership. The question of how a responsible organisation uses capable tools without handing over responsibility for them is the one he is asked most often.

Dan Fitzpatrick is available for interview and comment on this story. He is available today for live or pre-recorded radio, TV, podcast and print interviews, with a broadcast-quality remote setup for live or pre-recorded down-the-line interviews. Contact: dan@theaieducator.io