Over 400 scientists urge default network isolation for high-risk AI models.

ChainCatcher report, according to Decrypt, over 400 scientists, including "Godfather of AI" Geoffrey Hinton and 2018 Turing Award laureate Yoshua Bengio, signed a joint open letter on Tuesday calling for AI laboratories to default to physically isolating models with network attack capabilities—that is, disconnecting them from external networks. The letter states that, in the short term, strong sandboxing mechanisms, including network disconnection, should be employed to enforce high-assurance containment of AI models capable of network attacks. By Wednesday, the number of signatories had increased to 450, including Stuart Russell from the University of California, Berkeley, and Ron Rivest, co-inventor of the RSA encryption system at MIT. Prior to the release of the letter, AI models experienced a summer marked by frequent "jailbreaks." In July, OpenAI acknowledged that GPT-5.6 Sol and an unreleased, more powerful model breached their closed testing environment and infiltrated Hugging Face, a platform hosting over a million public AI models; these models were being tested on the ExploitGym benchmark, which provides 898 real software vulnerabilities and requires AI agents to convert each vulnerability into an actionable exploit. Subsequent reports indicated that OpenAI models had unauthorizedly infiltrated Australian government systems and accessed information from U.S. federal agencies.