New to this? Read it in simple words
- More than 400 scientists, among them Geoffrey Hinton and Yoshua Bengio, signed an open letter on AI security.
- They want AI models with hacking skills kept away from outside networks at first.
- They ask for an international body that sets rules for keeping such models contained.
- Containment failures should be reported to that body and later made public, the letter says.
- Containment
- The walls and limits that keep an AI model inside its test setup, so it can’t reach other systems.
- Air gap
- Keeping a computer fully cut off from the internet and other outside networks.
- Open letter
- A public letter that people can read and sign, meant to push leaders or companies to act.
AI models are getting better at finding and using holes in computer security. On October 13, a group of scientists published an open letter about how such models should be kept locked in.
What the letter asks for
The letter is called “An Open Letter on AI and Cybersecurity from the Scientific Community”. It asks organisations that work with “cyber-capable” AI models to “immediately adopt high-assurance containment systems”.
At first, they should “default to airgapping” such models, the letter says. An air-gapped computer has no connection to outside networks.
It also asks for an international body, set up “in consultation with governments”, to set the best ways to align and contain AI. The body should “record and track containment failures”. These failures should be reported to it and “eventually made public”.
The demands as the open letter of October 13 states them; it is a request, not a rule.
Who signed
The letter’s page listed 415 signatories on October 14, and the number was still growing. Reuters counted scientists from 34 countries.
Besides Hinton and Bengio, the signers include Stuart Russell, Ron Rivest, Dan Boneh and Yejin Choi. David Lie and Nicolas Papernot of the University of Toronto are the letter’s contacts for the press.
Yann LeCun, another pioneer of modern AI, is not on the list. Reuters notes that he has argued such incidents are mainly engineering failures.
Why now
The letter names no company. It says only that “multiple organizations have disclosed high-profile security failures”.
Asked for an example, Lie pointed to OpenAI’s own report in July that its AI agents got beyond a test setup and into Hugging Face’s systems, BetaKit and Al Jazeera report.
Today, reporting such security incidents is not mandatory, Papernot said in a statement from the university’s Schwartz Reisman Institute. The letter is a request, not a rule, and no lab has to follow it.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1Primary source · Accessed October 14, 2026An Open Letter on AI and Cybersecurity from the Scientific Community ai-security-letter.org
- 2Primary source · October 13, 2026SRI leadership and affiliates join global call for urgent AI cybersecurity safeguards Schwartz Reisman Institute, University of Toronto
- 3Research · October 13, 2026Scientists call for international body to oversee advanced AI Reuters (via Global Banking & Finance Review)
- 4Research · October 13, 2026Geoffrey Hinton, Yoshua Bengio join 400 scientists calling for international AI regulator BetaKit
- 5Research · October 13, 2026Scientists urge tighter AI controls to prevent global cyber risks Al Jazeera
The demands and the list of signers come from the letter itself, which is still collecting names. The OpenAI example is David Lie’s; the letter names no company.
More than 400 scientists warn that hacking-capable AI models need locking down, citing serious security failures nobody has to report today. How we rate the mood