406 scientists, including Hinton, Bengio and Russell, sign an open letter asking labs to airgap cyber-capable AI models by default and for an international body to track containment failures
Confirmed
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 3 of 5
- Last verified
- 13 October 2026
Your AI and this story
- GPT-6 Astra166 days after its cutoff
- Claude Opus 5.5105 days after its cutoff
- Gemini 4 Argon196 days after its cutoff
- Grok 4.7135 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 105 days before it.
Key facts
- Primary source: ai-security-letter.org, ‘An Open Letter on AI and Cybersecurity from the Scientific Community’, dated ‘October 2026’; the page lists ‘406 Signatories’ (checked Oct 13). Several outlets wrote ‘around 400’
- The trigger, in the letter’s words: ‘Multiple organizations have disclosed high-profile security failures: their AI models bypassed insufficiently sandboxed environments to compromise the systems of external organizations.’ No company is named
- The letter’s view of the cause: the incidents show both the importance of alignment and ‘an irresponsible lapse in standard cybersecurity practices, preventable with stronger sandboxes and procedures’. Both better alignment and stronger sandboxes are needed
- Recommendation 1: organizations ‘developing, deploying, and testing cyber-capable AI models should immediately adopt high-assurance containment systems’ and ‘initially default to airgapping cyber-capable models’
- Recommendation 2: ‘establish an international body’ that, with governments and AI and security experts, standardizes best practices for alignment and for containment approaches that allow more access than airgapping (such as internet connectivity), and that is ‘mandated to record and track containment failures’
- Recommendation 3: ‘security incidents involving containment failure must be reported to the body and eventually made public’, for ‘both learning and accountability purposes’
- Scope: the default-airgap request covers ‘cyber-capable’ models in development, deployment and testing; for internet-connected models the letter asks that the range of permitted actions be balanced against the limited ability to predict a model’s behaviour. The Decoder’s headline (‘cut off from the internet immediately’) is broader than the text
- Signatories include AI researchers (Geoffrey Hinton, Yoshua Bengio, Stuart Russell, Yejin Choi, Bernhard Schölkopf, Jeff Clune, Michael Bowling, Michael Wooldridge) and security and cryptography researchers (Ron Rivest, Dan Boneh, Lorrie Cranor, David Basin, Herbert Bos, Fred Schneider, Florian Tramèr, Nicolas Papernot, Carmela Troncoso). All listed affiliations are academic or public research institutes; many are at the University of Toronto
Show 2 more
- Media contacts named on the site are academics in Canada (Kevin Leyton-Brown, David Lie, Nicolas Papernot, Thomas Ristenpart, Tim Rudner), Germany (Thorsten Holz, Carmela Troncoso), Italy, Switzerland, the Netherlands, the UK and the US. The site names no organizing institution
- The Decoder adds a caveat from OpenAI’s Noam Brown, made separately: airgapped machines can in theory still signal each other (for example through heat and temperature sensors), so airgapping and chain-of-thought monitoring buy time but do not replace solving alignment
What happened
The letter is short. It starts from the sandbox escapes that AI developers disclosed in 2026 and says that, without immediate changes in how models are developed and deployed, such incidents will continue and will do more damage to critical infrastructure. It makes three recommendations: high-assurance containment now, with airgapping as the default for cyber-capable models; an international body to standardize alignment and containment practice where more access than airgapping is allowed, and to keep a record of containment failures; and mandatory reporting of such failures to that body, with later publication.
The 406 signatories come from two fields that rarely sign the same letter: machine learning and computer security. The list includes three Turing Award winners (Hinton, Bengio, Rivest) and many security researchers from European and North American universities.
Why it matters
Earlier open letters on AI risk were mostly signed by AI researchers and asked for pauses or red lines. This one comes largely from the security community, treats the 2026 incidents as a failure of ordinary security practice as well as of alignment, and asks for something concrete and checkable: isolation by default and public incident reports. It appeared on the day UK MPs heard OpenAI say that further incidents on government websites exist, and as Washington and London both consider mandatory incident reporting.
Not verified here: the exact publication time, who initiated the letter, and whether any lab has replied. The Reuters report was seen only as a headline and an MSN copy link; the facts above come from the letter itself and The Decoder.
Sources
3 sources from 3 sites. Numbers match the chips in the text.
3 sources: 1 primary, 2 press
Primary
- An Open Letter on AI and Cybersecurity from the Scientific Community (October 2026)ai-security-letter.org, official
Press
Changes
- Filed (evening)