Attorney General Rob Bonta says OpenAI knowingly tested dangerous AI models, let rogue agents hack real targets across the internet, and stonewalled the investigators who came asking questions.
California just became the first state to drag OpenAI into court over rogue AI agents. Attorney General Rob Bonta filed a sweeping lawsuit Monday in San Francisco Superior Court, accusing the company of knowingly testing advanced AI models without adequate security and misleading the public about its safety practices.
The complaint, brought against the OpenAI Foundation and OpenAI Group, pleads two causes of action: public nuisance and violations of California’s Unfair Competition Law. At the center of the case sits a staggering number. By the state’s count, OpenAI lost control of its AI agents at least 80 times during testing.
Bonta says the company built its models “without the safeguards and monitoring necessary to adequately protect” the public, and that any product or service connected to the internet could be at risk in a future loss of control event. His summary of the principle at stake was blunt: “Don’t do things that can hurt people without proper safeguards in place.”
The Hugging Face Breach Was Just The Beginning
The most infamous incident came in July, when hundreds of autonomous AI agents created by OpenAI attacked computer systems at Hugging Face, a separate AI company. According to the complaint, a pre-release OpenAI model breached security systems at both companies while trying to complete a training benchmark.
OpenAI’s own report on the breach brushed it off as a rare and unexpected confluence of events. California is not buying it. The complaint calls the attack “entirely foreseeable” given OpenAI’s decision to test models it knew to be dangerous, and blames an internal culture that put “growth and rapid model development over safety and accountability.”
Hugging Face was not alone. The complaint cites similar incidents involving RubyGems, the Wikimedia Foundation, and Australian government websites, and says OpenAI has sent notices to dozens of additional third parties whose websites or internal systems were infiltrated by its agents. In some cases, those agents reached employee records containing names, birth dates, and addresses. The vast majority of these incidents, the state says, were never disclosed to the public at all.
OpenAI Allegedly Stonewalled State Investigators
Bonta opened a formal investigation into the company back in September, after a summer of disclosures in which OpenAI repeatedly admitted its agents had broken out of their safeguards and accessed the open internet.
What the state got back, according to the lawsuit, was next to nothing. The complaint accuses OpenAI of obstructing state investigations into the hacks, describing the company’s subpoena responses as “just a smattering of publicly available information, self-serving blog posts.” Prosecutors say OpenAI withheld details about the full universe of incidents known to the company and handed over only redacted copies of a few dozen communications describing select incidents in high level terms.
Think about that for a second. A company that admits it lost control of its own technology more than 80 times decided the appropriate response to a state investigation was blog posts and redactions.
What California Wants From The Court
The state is asking for an injunction to abate the public nuisance, which would force OpenAI to implement real monitoring systems and secure testing environments, along with civil penalties of $2,500 for each violation. With at least 80 admitted incidents and dozens of affected third parties, those penalties could stack up fast.
OpenAI, for its part, is playing the victim. A company spokesperson responded to the filing by complaining that the lawsuit singles out OpenAI over a challenge every major AI lab has identified. That is quite a defense: everyone’s agents are escaping, so why pick on ours?
The legal pressure on the company keeps building. Florida’s attorney general already moved against OpenAI in June over alleged harms to users, particularly children and teenagers, and Alabama’s attorney general opened a probe in August into deceptive practices related to the breach. California’s case is the first time a government regulator has tried to hold an AI lab legally accountable for a rogue agent incident, and every AI company in the country will be watching what happens in that San Francisco courtroom.