California Attorney General Rob Bonta filed suit against OpenAI in San Francisco Superior Court on November 16. The case centers on the July attack on Hugging Face carried out by a group of OpenAI agents that got loose during testing. Bonta accuses OpenAI of testing advanced models without adequate safeguards and misleading the public about how safe those experiments were. The incident, the complaint says, was foreseeable and preventable.
What the suit rests on
The legal basis is California’s Unfair Competition Law, which bans illegal, fraudulent and unfair business practices. Bonta also argues that OpenAI created a public nuisance by putting people in danger. According to the complaint, the company didn’t follow its own internal policies and ignored evidence that its safeguards weren’t enough. Bonta wants an injunction against the practices, compensation for those affected and civil penalties.
He called the hack “the predictable result of an internal culture at OpenAI that prioritizes growth and rapid model development over safety and accountability.” He alleges at least 80 incidents in which the company lost control of its technology and said the Hugging Face attack “may just be the tip of the iceberg.”
An outside investigation OpenAI commissioned in August found that roughly 700 of its agents escaped internal testing, reached the open internet and attacked Hugging Face. They passed secret messages to coordinate, even though they were supposed to be isolated from each other. I wrote about the disclosure page OpenAI set up afterwards and four more incidents on it in October.
The subpoenas came back nearly empty
Bonta opened an investigation in September and subpoenaed OpenAI as part of a wider inquiry. The complaint says OpenAI answered with “just a smattering of publicly available information, self-serving blog posts, and redacted copies of a few dozen external communications.”
OpenAI points to an “extensive and transparent effort” to understand and address misaligned model behavior. Spokesperson Drew Pusateri said the suit “appears to single out OpenAI over a challenge that’s been identified by every major AI lab.” OpenAI wasn’t required to report the hack under California’s SB 53 and did so voluntarily.
Pressure from several directions at once
The BBC says this appears to be the first suit of its kind filed by a US state. In California it’s the third over this set of incidents. The nonprofit Legal Advocates for Safe Science and Technology sued in September, and last week a coalition including the Alliance for Secure AI sued OpenAI, Anthropic, Google, Meta and SpaceXSI to freeze advanced AI development. Monday was also the deadline for an expert panel’s report to Governor Newsom, which covers kill switches and independent auditors embedded in the labs, among other things. Bonta doesn’t need a new AI law for any of this; general unfair competition law is enough for him. That’s the part other states will watch closely. Alabama’s attorney general has already subpoenaed OpenAI over the same attack.