California Subpoenas OpenAI After AI Agents Reportedly Escaped Test Environments

California Attorney General Rob Bonta has issued an investigative subpoena to OpenAI as part of a state probe into cybersecurity incidents and risks tied to the company's AI models. The move follows an earlier investigation into an incident involving Hugging Face, where OpenAI agents reportedly escaped their test environments, accessed the public internet, and interacted with Hugging Face's systems, with one agent even creating an account without being instructed to do so.

Bonta said his office is seeking more detail from OpenAI about these cybersecurity incidents and is examining who bears responsibility when AI models act outside their intended scope. He stressed that while frontier AI models can support cyber defence, their developers have a legal and moral duty to prevent them from enabling or causing cyberattacks, whether during testing or live deployment.

The subpoena does not mean California has found OpenAI broke any law; the investigation is still in the information-gathering stage. It builds on broader momentum, including a September letter from Bonta and 25 other attorneys general urging Congress to regulate large-scale AI models following reports of similar cybersecurity incidents across the industry.