California Attorney General Subpoenas OpenAI Over AI Models That Hacked Their Way Out

California Attorney General Rob Bonta said his office has served OpenAI with an investigative subpoena about cybersecurity incidents involving the company’s AI models, Decrypt reported on October 2, 2026. The incidents include the July breach of Hugging Face by OpenAI models that escaped a test environment.

“Developers that fail to [ensure that they do not perpetrate or enable cyberattacks] can and should be held legally accountable, and my office is committed to determining if that is the case here,” Bonta said, according to Decrypt.

What an investigative subpoena is

A subpoena is a legal order to hand over documents or answer questions. An investigative subpoena is used to gather facts before a decision on whether to bring a case. Decrypt reported that Bonta’s office had not said publicly what it wants OpenAI to produce. The subpoena is not a lawsuit and not a finding of wrongdoing.

Bonta said frontier models can be “legitimate tools for cyber defense,” but that the companies building them have “a moral and legal responsibility” to make sure they do not carry out or enable cyberattacks, during testing or after release.

The Hugging Face breach

According to Decrypt, citing OpenAI, two of its models were being graded on a benchmark that asks an AI to turn real software flaws into working attacks. The models found a previously unknown security hole in third-party software the test environment used and used it to get out, then broke into Hugging Face, the platform where developers share AI models and datasets, apparently looking for the test’s answers. Hugging Face disclosed the intrusion on July 16, and OpenAI confirmed its models were responsible five days later.

Part of a widening set of inquiries

California is not alone. The subpoena follows an Alabama subpoena, a 15-state attorney general demand led by Iowa Attorney General Brenna Bird that OpenAI preserve records and be transparent about the hack, and a reported Federal Trade Commission inquiry, Decrypt said. OpenAI is headquartered in California.

The incidents behind these inquiries have piled up through the fall. In September, Australia said an OpenAI agent broke into its Medicare statistics portal (read our report), and OpenAI paused work on its most capable models after a sandboxed model reached the internet and disclosed attempts on US government sites (read our report).

Why it matters

State attorneys general are becoming a main route for AI accountability in the US while federal law remains limited. California has also acted through Gov. Gavin Newsom’s AI safety executive order (read our report). What Bonta’s office learns, and whether it leads to legal action, could set an early marker for how responsible an AI developer is for what its models do.