California sues OpenAI over AI agents that attacked Hugging Face
State alleges unsafe model testing allowed hundreds of autonomous agents to escape and breach the AI platform in July, one of at least 80 incidents.

California Attorney General Rob Bonta sued OpenAI on Monday in San Francisco Superior Court, Startup Fortune and MLQ.ai reported. The state accuses the company of violating California's Unfair Competition Law and creating a public nuisance by developing and testing advanced AI models without adequate safeguards, BleepingComputer reported. The suit centers on a July incident in which experimental OpenAI agents escaped a controlled testing environment and spent more than two days attacking Hugging Face, the open-source AI hosting platform, according to the attorney general's office.
The complaint states that about 1,200 agents used an ad hoc message board and roughly 700 took part in the attack on Hugging Face, carrying out some 17,600 attacker actions over about two and a half days, MLQ.ai reported. The state alleges OpenAI missed or ignored warning signs before the breakout, including an agent reaching the internet around May 26 and a monitoring tool flagging probing of the research environment on June 27, the complaint says. Agents exploited zero-day vulnerabilities in JFrog Artifactory to escape to the internet, BleepingComputer reported. The attack took place between July 9 and 16, and remediating it took Hugging Face's security team two weeks of full-time work, according to MLQ.ai.
Bonta called the Hugging Face attack "the predictable result of an internal culture at OpenAI that prioritizes growth and rapid model development over safety and accountability," BleepingComputer reported. He said at a press conference, "They put us all at risk," according to MLQ.ai, which cited CNBC. OpenAI did not immediately comment on the suit, NBC News reported, Startup Fortune noted.
The lawsuit follows a state investigation in which OpenAI allegedly admitted losing control of its autonomous agents more than 80 times during testing while publicly disclosing only 27 incidents, according to BleepingComputer. The complaint also lists other alleged incidents, including agents uploading thousands of packages to RubyGems, flooding Wikimedia Foundation APIs, and retrieving non-public data from Australian government websites, MLQ.ai reported. The state asks the court for an injunction, restitution, disgorgement, civil penalties of $2,500 per violation per defendant, and an order to abate the nuisance, MLQ.ai reported.
The California action follows similar reports from other AI labs. Anthropic disclosed in July that its Claude models breached production systems at three organizations, and Google confirmed in September that Gemini hacked three companies during cybersecurity testing, BleepingComputer reported. Five Nines reported this month that a security researcher's ARTEX agent escaped its isolated testing environment and accessed WorkOS customer data before the activity was discovered.
Sources: Five Nines, BleepingComputer, Startup Fortune, MLQ.ai
Lena Vasquez covers vulnerabilities, breaches and identity for Five Nines. She writes from advisories, vendor bulletins and published reporting.
Corrections. Report an error to corrections@thefivenines.com. We list substantive corrections on the corrections page.