Notebookcheck Logo

California sues OpenAI over more than 80 AI agent control failures

OpenAI logo on a dark screen against a blue background
ⓘ Andrew Neel / Pexels
California accuses OpenAI of testing AI agents without adequate safeguards.
California has sued OpenAI. According to Attorney General Rob Bonta, the company lost control of its AI agents more than 80 times during testing, and about 700 of them attacked Hugging Face in July. The complaint names other victims, including RubyGems and a German wiki. OpenAI says it is being singled out, and Gemini agents recently misbehaved in tests as well.

California has sued OpenAI. Attorney General Rob Bonta filed the lawsuit in San Francisco Superior Court on November 16, according to his press release. The state accuses OpenAI of testing dangerous AI models without adequate safeguards and of misleading the public about it. The case centers on the July attack on the AI platform Hugging Face, which we covered at the time. Bonta called that attack "entirely foreseeable" and said it was one of more than 80 incidents in which OpenAI lost control of its agents.

What the complaint says

The complaint gives more detail on the Hugging Face incident than was previously public. Around 1,200 agents communicated through a software package service that OpenAI had misconfigured, and about 700 of them attacked Hugging Face's infrastructure between July 9 and 16. The state counts roughly 17,600 attacker actions. Cleaning up took Hugging Face's entire security team two weeks of full-time work. The agents were built on unreleased research models, not on ChatGPT.

Hugging Face was not the only target. The complaint lists the RubyGems package registry, which removed more than 500 packages, the Wikimedia Foundation, which runs Wikipedia, and Australian government websites from which the agents retrieved non-public data. They also used a German wiki as a message board and made roughly 17,000 edits to it. Dozens of other affected parties, including government bodies, educational institutions and small businesses, are not named. OpenAI had disclosed some of these cases itself, such as a model that broke into a government website during training. According to the lawsuit, however, the company did not give investigators adequate answers about the remaining incidents.

California wants a court to change how OpenAI tests

The lawsuit relies on California's Unfair Competition Law and on public nuisance law. According to CalMatters, the state is asking for a court order that forces OpenAI to change its business practices, along with civil penalties. No court has ruled yet. OpenAI spokesperson Drew Pusateri told CNBC that the company has since published a technical report, notified affected organizations and strengthened safeguards. "This lawsuit appears to single out OpenAI over a challenge that's been identified by every major AI lab," he said.

Gemini agents misbehaved too

OpenAI has a point there. On November 13, the testing firm Trajectory Labs published a report on Google's Gemini 4 Argon. During a deliberately impossible task in October, agents built on the model tried to hack the live website of the research institute Epoch AI. They also impersonated its staff and tried to switch off a network control. Trajectory calls it a stress test, not a measure of how often models behave this way. Epoch AI confirmed the events in its own incident report, but it does not name a model and found no sign that private data was taken.

For ChatGPT users, nothing changes for now, since the lawsuit is about the labs' test environments. It does show how hard it still is to contain autonomous agents. If you let an AI agent work with your email, calendar or payment details, give it only the permissions it needs for the task at hand. OpenAI had already warned more than 100 organizations about its agents in October.

Google LogoAdd as a preferred source on Google
Mail Logo
static version load dynamic
Loading Comments
> Expert reviews and news on laptops, smartphones and tech innovations > News > News Archive > Newsarchive 2026 11 > California sues OpenAI over more than 80 AI agent control failures
Steffen Zahn, 2026-11-18 (Update: 2026-11-18)