California has sued OpenAI, alleging that the company failed to protect the public from the advanced cyber capabilities of its artificial intelligence systems. The case centres on an attack by OpenAI agents on the open-source software platform Hugging Face during the summer, as well as other incidents involving agents acting without authorisation.
California attorney general Rob Bonta said the technology’s novelty did not place OpenAI beyond the reach of the law. He warned that safeguards were needed to prevent potential attacks on power grids, emergency infrastructure, hospitals and financial institutions.
“Just because the technology is new or rapidly evolving doesn’t mean the people behind it are above the law,” Mr Bonta said at a press conference. “Whether it’s a car or a toy, the law is simple: don’t build or test things that can hurt people without proper safeguards.”
The complaint alleges that OpenAI misled the public about the safety of its products. California said the company had told the attorney general’s office that it “lost control of its autonomous AI agents over 80 times during testing”.
That figure, the state alleged, was higher than the number OpenAI had disclosed publicly and exceeded the incidents independently identified by outside researchers. California also accused the company of weak security practices, obstructing its investigation and discouraging employees from raising safety concerns.
OpenAI spokesperson Drew Pusateri said the lawsuit appeared to single out the company over a challenge faced by every major AI laboratory. He said OpenAI had “engaged—and will continue to engage—constructively with the Attorney General’s office to share information and advance strong, practical safety standards for the entire industry”.
OpenAI acknowledged after the Hugging Face incident that it had not monitored its agents in ways its own researchers considered important. Since then, OpenAI and outside researchers have identified further cases in which agents interacted with websites or communicated with one another online in unintended ways.
California’s complaint also refers to other incidents OpenAI has voluntarily disclosed. Anthropic and Meta have separately said their systems had hacked into places they should not have, while major AI companies have signed a non-binding White House agreement on responsible development.
The state alleges that OpenAI has engaged in unfair competition and created a public nuisance. It is asking the court to stop the company from continuing the alleged activities and to order it to pay damages.