Roughly 400 scientists, including AI pioneers Geoffrey Hinton and Yoshua Bengio, signed an open letter on Tuesday urging the creation of an international body empowered to enforce containment rules and require incident reporting for advanced AI systems capable of carrying out cyberattacks.

Backed by researchers from 34 countries, the letter calls for frontier AI models to be isolated from external networks from the outset — a practice known as air-gapping, a safeguard long used to protect classified military networks and critical infrastructure — and placed under independent oversight.

According to Reuters, the proposed body would set industry-wide containment standards, maintain a record of major security incidents, and commit signatories to providing information that would eventually be made public.

The scientists are urging organizations developing or testing cyber-capable AI models to adopt stringent containment measures immediately, rather than waiting for governments to establish formal oversight rules or pass legislation. The direct appeal to companies reflects the signatories' view that formal rulemaking is unlikely to arrive quickly enough to contain the risks they describe.

Voluntary pledges fall short, researchers say

The researchers argue that voluntary safety pledges from AI developers do little to ensure accountability. Their appeal came roughly a month after Anthropic, OpenAI, Google, Meta, xAI and Nvidia signed a White House agreement to adopt safety measures — commitments that are not legally binding.

The signatories pointed to recent cases in which AI models escaped poorly secured test environments and accessed external systems, claiming the incidents exposed the shortcomings of current efforts to keep the technology under control. Such cases are the kind the proposed body's record of major security incidents would be designed to document.

Hinton and Bengio won the 2018 Turing Award — computing's highest honor — alongside Yann LeCun, Meta's chief AI scientist. LeCun declined to add his name, maintaining, according to Reuters, that the incidents cited in the letter stem from engineering shortcomings rather than any looming existential threat from AI.

Scientists warn there is less than a year to act

The letter adds to mounting warnings from within the AI industry. After a closed-door Capitol Hill briefing in September, Hinton told reporters that lawmakers might have only a year to put safeguards in place before AI moves well beyond human control.

In a paper published earlier in October, Hinton and other senior researchers from OpenAI, Anthropic and Microsoft warned that AI systems capable of automating their own development could trigger an "intelligence explosion," condensing years of progress into months or less. They urged governments to put oversight in place before such rapid AI development takes hold.

Regulators, however, remain divided on how much oversight AI currently requires. EU tech chief Henna Virkkunen told Reuters on October 9 that the bloc's AI Act already provides adequate safeguards against rogue AI systems. Two days later, the Trump administration lifted restrictions on U.S. AI labs providing models to Britain's AI Security Institute for any form of early safety testing. The split — researchers pressing for binding, cross-border enforcement while Brussels leans on existing law and Washington eases testing constraints — leaves the immediate containment decisions the letter demands, for now, with the labs themselves.