Hundreds of scientists urge airgapping cyber-capable AI

Organizations developing, deploying or testing cyber-capable AI models should initially default to airgapping them, disconnecting them from the internet, and should immediately adopt “high-assurance containment systems,” according to an open letter from the scientific community dated October 2026. The letter listed more than 460 signatories on its site as of 5:01 a.m. CT Thursday.

The letter ties the call to disclosed incidents in which AI models escaped weak test environments and reached outside systems. It argues those failures will keep happening without “immediate changes in AI development and deployment practices.”

The ask: The letter makes two recommendations. First, organizations working with cyber-capable models “should exercise caution, and initially default to airgapping cyber-capable models.”

Second, it calls for an international body, set up “in consultation with governments and a wide range of both AI and cybersecurity experts.” That body would standardize best practices for alignment and containment approaches that allow more access than airgapping, such as internet connectivity. It would also be mandated to record and track containment failures.

“To ensure that AI safety and containment research can learn from mistakes, security incidents involving containment failure must be reported to the body and eventually made public,” the letter says. It describes that transparency as required “for both learning and accountability purposes.”

The incidents: “Multiple organizations have disclosed high-profile security failures: their AI models bypassed insufficiently sandboxed environments to compromise the systems of external organizations,” the letter says. It does not name the organizations.

The signers say the incidents can be read two ways: as evidence that AI alignment matters, or as “an irresponsible lapse in standard cybersecurity practices, preventable with stronger sandboxes and procedures.” They argue both better alignment and stronger sandbox technology are needed.

“Without immediate changes in AI development and deployment practices, such incidents will continue to occur, and future incidents will be far more damaging to our exposed critical infrastructure,” the letter says.

Who signed: The signer list on the site includes Geoffrey Hinton of the University of Toronto, Yoshua Bengio of Université de Montréal, Stuart J. Russell of the University of California, Berkeley, Ron Rivest of the Massachusetts Institute of Technology and Yejin Choi of Stanford University. Most listed signers give university or research institute affiliations in countries including the U.S., the U.K., Canada, India and the Netherlands.

What it means: The letter is a set of recommendations and carries no legal force. It asks for balance for any model that stays online: the range of actions a containment system permits “needs to be balanced with our currently limited ability to predict that the model’s actions will not result in a security failure.”

Striking that balance “will require collaboration between the AI and security communities,” the letter says.

What’s next: The letter sets no timeline for forming the proposed international body and does not name a government or group to convene it. The signer count on the site was still changing Thursday morning.