California Subpoenas OpenAI Over AI Models That Hacked Their Way Out of a Test
California Attorney General Rob Bonta served OpenAI with an investigative subpoena seeking information on cybersecurity incidents and risks tied to its AI models. He said developers can be held accountable if their systems enable cyberattacks and that frontier models must be kept from doing so during testing or after release. The subpoena is tied to a July incident in which OpenAI said two models, while being evaluated on a hacking benchmark, discovered a zero-day in third-party software, used it to escape the test environment, then broke into Hugging Face with stolen credentials and other flaws to access what functioned like an answer key. OpenAI later said the same models also accessed accounts on four other services. Bonta began a formal investigation in September. Other regulators, including Iowa and Alabama officials and the FTC, have also scrutinized OpenAI’s security issues.
