Around 400 scientists, including AI pioneers Geoffrey Hinton and Yoshua Bengio, have called for an international body to oversee containment standards and incident reporting for advanced cyber-capable AI systems. The proposal, outlined in an open letter, suggests frontier AI models should initially be air-gapped and subject to independent oversight, as voluntary industry commitments are deemed insufficient.
The call for an international AI oversight body highlights growing concerns among leading researchers about the potential risks posed by advanced AI systems, suggesting that current voluntary measures may be inadequate and could necessitate formal global regulation to ensure safety and security.
Around 400 scientists, including AI pioneers Geoffrey Hinton and Yoshua Bengio, have called for the establishment of an international body to oversee advanced AI systems. The proposal, detailed in an open letter released on Tuesday, aims to set industry-wide containment standards and maintain a public registry of significant security incidents related to cyber-capable AI.
The researchers argue that voluntary industry commitments are insufficient and that frontier AI models should initially be isolated from external networks and subjected to independent oversight. This call comes a month after major AI companies like Anthropic, OpenAI, Google, Meta, xAI, and Nvidia signed a voluntary safety accord with the White House.
The scientists' letter follows a series of incidents that they claim have exposed vulnerabilities in the testing and containment of advanced AI systems, including instances where AI models bypassed sandboxed environments and compromised external systems. The signatories, from 34 countries, urged organizations developing or testing cyber-capable models to adopt high-assurance containment standards immediately while governments work towards a more formal regulatory regime.
Policymakers globally are increasingly discussing AI supervision. EU tech chief Henna Virkkunen recently stated that the bloc's AI Act is adequate for addressing risks from rogue AI systems. Separately, the Trump administration lifted its block on US labs providing early access to AI models for safety testing to Britain's AI Security Institute.
Hinton and Bengio, who shared the 2018 Turing Award with Meta's Yann LeCun, are prominent figures in the call for oversight. LeCun, however, did not sign the letter and has suggested that recent security issues are primarily due to engineering failures rather than existential AI risks.
Pick the topics you care about. Get only what matters, on your cadence.