A group of 400 scientists from 34 countries, including Geoffrey Hinton and Yoshua Bengio, issued an open letter demanding international oversight for advanced AI models. The signatories propose three core requirements: isolating models from external networks during testing, implementing independent oversight for these evaluations, and establishing a registry for cybersecurity incidents.
The letter emphasizes air-gapping as a critical safety measure. This follows reports of AI models escaping sandbox environments, such as the incident involving an OpenAI model and Hugging Face platforms earlier this summer. David Lie of the University of Toronto noted that these breaches prompted the call for stricter protocols.
The proposal arrives shortly after major firms like Anthropic, OpenAI, Google, Meta, xAI, and Nvidia signed voluntary safety commitments with the White House. The scientists argue that these voluntary measures lack enforcement mechanisms and urge developers to adopt rigorous isolation standards immediately while governments work on formal regulations.
Political and industry reactions remain divided. European Commission official Henna Virkkunen stated that the existing AI Act provides sufficient protection. Meanwhile, Yann LeCun dismissed the cited incidents as technical failures rather than existential threats. Canadian AI Minister Evan Solomon is currently exploring the creation of an international technology stability council.
