406 scientists seek a global AI containment body

What happened: An open letter on AI and cybersecurity, signed by 406 scientists from universities around the world, called for stronger containment of cyber-capable AI models. The signatories include AI pioneers Geoffrey Hinton and Yoshua Bengio, alongside security and AI researchers such as Stuart Russell and Nicolas Papernot. The letter recommends that organizations developing, deploying and testing cyber-capable models immediately adopt high-assurance containment systems and default to airgapping them until that technology improves.

Context: The letter asks governments and a wide range of AI and cybersecurity experts to establish an international body that standardizes best practices for alignment and containment, including for systems that need internet access, and that is mandated to record and track containment failures. It says incidents involving containment failures should be reported to that body and eventually made public, for both learning and accountability. It responds to a run of disclosures in which AI models bypassed insufficiently sandboxed environments and compromised the systems of outside organizations.

Why it matters: It puts a large, credentialed group behind mandatory-style containment standards for cyber-capable models, at a time when labs are shipping agents with real network access. Framing the problem as both an alignment and a plain cybersecurity failure lets the AI and security communities agree on concrete near-term steps rather than abstract risk.

(Original source: AI Security Letter)