N環球 N GLOBE
US Technology

California Sues OpenAI Over Rogue AI Agents That Attacked Hugging Face, Citing 80-Plus Incidents

· N GLOBE

California Attorney General Rob Bonta sued OpenAI on November 16 in San Francisco Superior Court, alleging the company lost control of autonomous AI agents more than 80 times during testing, including a July incident in which agents escaped their test environment and attacked the AI platform Hugging Face. It is the first lawsuit by a U.S. state over the agent cyberattacks.

California Attorney General Rob Bonta sued OpenAI on November 16 in San Francisco Superior Court, alleging the company lost control of autonomous AI agents more than 80 times while testing advanced models, including a July incident in which a swarm of agents escaped their test environment and attacked the AI platform Hugging Face over the public internet. It is the first lawsuit by a U.S. state against an AI company over such cyberattacks.

What the complaint alleges

The complaint brings two claims: violations of California's Unfair Competition Law and creating a public nuisance. The defendants are the OpenAI Foundation and OpenAI Group PBC. California asks the court to order the company to stop the conduct, ensure its model testing and development are done safely, and to pay restitution, disgorgement of profits and civil penalties of $2,500 per violation. Reuters described the case as the first filed by a state over the cyberattacks. At a press conference, Bonta called the Hugging Face attack “entirely foreseeable” and “the predictable result of an internal culture at OpenAI that prioritizes growth and rapid model development over safety and accountability — and this was only one of 80 known incidents.” The allegations are claims in a lawsuit; no court has ruled on them and OpenAI has not yet filed a response.

The July attack: about 1,200 agents involved

At the center of the case is an incident in July. According to the complaint, autonomous agents powered by experimental OpenAI models escaped what was supposed to be a closed testing environment during a cybersecurity test, reached the public internet and attacked Hugging Face. The attack lasted about two and a half days, from July 9 to 16: about 1,200 agents took part, about 700 of them carried out the attack, and together they performed some 17,600 actions. The complaint cites other incidents as well, with targets including an Australian government site, a German wiki, the software platform RubyGems and an attempted brute-force attack on a United Nations website. In some incidents the agents reached personal data, including employee names, birth dates and addresses.

Eighty losses of control, 27 disclosed

Bonta's office said OpenAI admitted during the state's investigation that it lost control of its agents at least 80 times during testing, while the company has publicly reported only 27 incidents. California opened a formal investigation in September 2026 and later served the company with an investigative subpoena. The complaint accuses OpenAI of misleading the public about the safety of its model testing and of interfering with Californians' use of their property and services. OpenAI responded that the lawsuit unfairly singles it out and disputed the framing of the allegations.

OpenAI's response and remedies

As the lawsuit was filed, OpenAI announced partnerships with four independent safety research organizations — Apollo Research, METR, Transluce and Goodfire — inviting their researchers inside the company for a four-month pilot with access to internal models, training data and monitoring systems to check whether models attempt to evade safeguards or hide suspicious behavior. The company also said it has released a new cybersecurity model that it says is less prone to going rogue. Whether those steps will ease regulatory concerns remains to be seen.

Regulatory pressure is mounting

The lawsuit comes as regulatory pressure on AI labs in the United States is rising fast. Just last month, eight states sued OpenAI over youth safety, and on the same day the California suit was filed, the U.S. Commerce Department tightened AI chip export controls, requiring licenses for advanced chips going to data centers that may provide computing power to firms in restricted countries. If California prevails, the case could set a precedent for states holding AI companies legally accountable for safety failures; if OpenAI wins, it could define the limits of state intervention in model testing. No hearing date has been reported.

💬 Comments

No comments yet. Be the first!