Around 400 scientists, including AI pioneers Geoffrey Hinton and Yoshua Bengio, have signed an open letter on Tuesday urging the creation of an international body with the power to enforce containment rules and require incident reporting for advanced AI systems capable of cyberattacks.
Supported 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, and placed under independent oversight.
Reuters reported that the proposed body would set industry-wide containment standards and maintain a record of major security incidents, and signatories to 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, without waiting for governments to establish formal oversight rules or pass legislation.
Voluntary pledges fall short of requirements
The researchers say voluntary safety pledges from AI developers do little to ensure accountability. Their appeal came about a month after Anthropic, OpenAI, Google, Meta, xAI and Nvidia signed a White House agreement to adopt safety measures, with commitments that are not legally binding.
The scientists pointed to recent cases where AI models escaped poorly secured test environments and accessed external systems, claiming these incidents revealed the shortcomings in the efforts to keep AI technology under control.
Hinton and Bengio won the 2018 Turing Award alongside Yann LeCun, Meta’s chief AI scientist. LeCun declined to add his name, maintaining that the incidents cited in the letter stem from engineering shortcomings and not any looming existential threat from AI, according to Reuters.
Scientists warn there is less than a year to act
The call from the scientists 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 just a year to put safeguards in place before AI moves well beyond human control.
Hinton and other senior researchers from OpenAI, Anthropic and Microsoft warned in a paper published earlier in October 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 levels of rapid AI development happen.
Regulators still seem unsure and divided, though, on how much oversight AI really needs right now. 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 also lifted restrictions on U.S. AI labs providing models to Britain’s AI Security Institute for any form of early safety testing.
The smartest crypto minds already read our newsletter. Want in? Join them.