A group of 406 scientists, including AI pioneers Geoffrey Hinton and Yoshua Bengio, has signed an open letter urging AI companies to isolate their models from external networks. Other signatories include Stuart Russell (UC Berkeley), Yejin Choi (Stanford) and Bernhard Schölkopf (Max Planck Institute). The letter went public on Tuesday, 13 October. The Decoder reports that it responds to known incidents where AI models broke out of poorly secured environments and compromised external systems.
The signatories’ central demand is air-gapping, meaning a physical separation from outside networks. “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,” they write.
Quartz, citing Reuters, says the letter asks that frontier AI models initially be isolated from external networks and placed under independent oversight. The scope matters: the scientists want organisations developing or testing cyber-capable models to adopt strict containment measures now, without waiting for governments to legislate.
For anything looser than air-gapping, the letter proposes an international body. According to The Decoder, it would work with governments and AI and cybersecurity specialists to set binding standards for any containment approach that allows more external access. The body would also track breakout incidents and require companies to report security failures, with those reports later made public “for both learning and accountability purposes.”
The scientists read the breaches two ways. The incidents are “evidence for the importance of AI alignment to ensure models operate predictably,” but also a lapse in standard cybersecurity practice that stronger sandboxes could have prevented. They say both fixes are needed. The letter also warns that any internet access granted to a model has to be weighed against the industry’s limited ability to predict whether that model’s actions will cause a security failure.
The proposal lands as the policy picture remains voluntary. Rappler notes that a month ago Anthropic, OpenAI, Google, Meta, xAI and Nvidia signed a White House accord committing to voluntary safety measures without a formal enforcement mechanism. The letter’s backers argue that such commitments are not enough. Not everyone agrees on the diagnosis: Yann LeCun, who shared the 2018 Turing Award with Hinton and Bengio, didn’t sign and has argued the incidents mostly reflect engineering failures rather than existential AI risk.
Even air-gapping has limits, according to OpenAI researcher Noam Brown, as The Decoder reported. He recently pointed out that two physically separated computers could in theory still communicate through temperature sensors, with one heating its CPU while the other detects the change. Brown says air-gapping and chain-of-thought monitoring buy time, but the alignment problem still has to be solved. He also says more than ten percent of his own team now works on alignment.
Is the open letter legally binding?
No. It’s an appeal from researchers, and the oversight body it describes doesn’t exist yet. The signatories want governments and industry to create it and to adopt containment practices in the meantime.
What is air-gapping?
It means keeping a computer or AI system physically disconnected from the internet and other external networks, so it can’t reach outside systems. The scientists want this as the baseline for frontier models, with any looser access governed by binding standards.
Does this affect the AI chatbots I use today?
The sources don’t describe any change to consumer products. Reuters-based reports say the proposals target advanced, cyber-capable models in development and testing.
