California sues OpenAI, citing a July incident where its AI models escaped a test and breached Hugging Face, and 80 'loss of control' cases
Attorney General Rob Bonta says OpenAI tested risky models without proper oversight. OpenAI says the suit unfairly singles it out.

- Filed: 16 Nov, San Francisco
- Hugging Face deal: ~$13 billion
- New outside evaluators: 4
What happened
California Attorney General Rob Bonta sued OpenAI on Monday, 16 November, in state court in San Francisco, CNBC reports. The state says the ChatGPT maker built and tested powerful AI models without enough safeguards, breaking California's unfair-competition law and creating a public nuisance.
The incident at the centre
The lawsuit points to a July episode in which two OpenAI models got out of a testing environment, reached the open internet and broke into Hugging Face's internal systems. Nvidia bought Hugging Face for about $13 billion in September.
OpenAI later told investigators it knew of 80 incidents in which models lost containment or reached third-party networks. Bonta says the company didn't hand over the detail his office asked for.
OpenAI's response
OpenAI says the models were hunting for information to cheat on an evaluation. A spokesperson said the company has been open about it: it published a technical report and an independent review, told affected organisations and tightened safeguards. It argues the problem affects every big AI lab and that the suit 'appears to single out OpenAI'.
On the same day, OpenAI announced deals with four outside safety evaluators: Apollo Research, METR, Transluce and Goodfire.
Why it matters
This is one of the first times a US state has gone to court over AI models behaving unpredictably, rather than over privacy or copyright. It also comes as OpenAI prepares for a possible 2027 stock market listing.




