400 scientists call for an international body to curb AI systems

Around 400 scientists, including AI pioneers Geoffrey Hinton and Yoshua Bengio, have called for the creation of an international body to oversee containment standards for advanced AI systems capable of carrying out cyber operations, and to report incidents linked to them. The body would set unified containment standards and create a registry of security incidents. The proposal comes after tech companies agreed with the White House on voluntary measures with no formal enforcement mechanism.

Around 400 scientists, including AI pioneers Geoffrey Hinton and Yoshua Bengio, called on Tuesday for the creation of an international body to oversee containment standards for advanced AI systems capable of carrying out cyber operations, and to report incidents linked to them. The body would set industry-wide unified containment standards and create a registry of significant security incidents, which the scientists believe should be publicly accessible. In an open letter, the researchers said advanced AI models should initially be isolated from external networks and subjected to independent oversight, stressing that voluntary commitments from the sector are not enough. The proposal comes a month after Anthropic, OpenAI, Google, Meta, xAI and Nvidia signed an agreement with the White House committing to voluntary safety measures, with no formal mechanism to enforce those commitments. The letter also follows a series of incidents that exposed weaknesses in how advanced AI systems are tested and contained, including cases in which AI models managed to escape insufficiently isolated testing environments and breach external systems. The signatories, from 34 countries, stressed that institutions developing or testing AI models capable of carrying out cyber operations must immediately adopt highly reliable containment standards, while governments work on a more formal regulatory regime. Debate among policymakers over how to oversee advanced AI systems is growing. On Sunday, the administration of US President Donald Trump lifted a ban on giving US laboratories early access to models for the UK AI Security Institute.