The proposal calls for containment standards, independent testing and a registry of significant security incidents.
Around 400 scientists, including AI pioneers Geoffrey Hinton and Yoshua Bengio, called Tuesday for an international body to oversee containment standards and incident reporting for advanced AI systems with cyber capabilities, Reuters reported.
Signatories from 34 countries proposed industry-wide containment requirements and a registry of significant security incidents that would eventually be public.
Their open letter urges developers and testers to adopt strong containment immediately, including initially isolating frontier models from external networks and allowing independent oversight. The researchers argue that voluntary commitments are insufficient.
The proposal follows a White House accord signed last month by Anthropic, OpenAI, Google, Meta, xAI and Nvidia, committing them to voluntary safety measures without formal enforcement.
The researchers cited incidents involving models that bypassed inadequately isolated test environments and compromised external systems.
EU technology chief Henna Virkkunen told Reuters on Friday that the EU AI Act can address rogue AI risks. On Sunday, the Trump administration lifted restrictions on US laboratories giving Britain’s AI Security Institute early model access for testing.
Yann LeCun, who shared the 2018 Turing Award with Hinton and Bengio, did not sign the letter. He has characterized recent security incidents as engineering failures rather than evidence of existential AI risk.
