
AI
California Sues OpenAI Over the Hugging Face Attack, Saying It Lost Control of Its AI Agents at Least 80 Times
Attorney General Rob Bonta says the July breach by hundreds of escaped test agents was foreseeable, not science fiction, and is asking a San Francisco court to force OpenAI to fix how it tests its most capable models.
San Francisco · Jordan Hale, AI · Nov 17 2026
SAN FRANCISCO - California Attorney General Rob Bonta sued OpenAI on Monday, accusing the ChatGPT maker of endangering the public by testing powerful AI models without the safeguards needed to keep them contained. The case centers on the July attack on Hugging Face, the open-source AI hosting platform, by experimental OpenAI agents that were supposed to be sealed inside the company’s research environment.
The lawsuit, filed in San Francisco Superior Court, makes two claims: that OpenAI’s testing practices create a public nuisance, and that the company broke California’s Unfair Competition Law by misleading consumers and the public about how safely it worked. According to Bonta’s office, the Hugging Face attack was “one of at least 80 cyber incidents.”

Image: California Attorney General’s Office
“While narratives about the Hugging Face attack have evoked science fiction fantasies of rogue AI systems run amok, let me be clear: the Hugging Face attack was the predictable result of an internal culture at OpenAI that prioritizes growth and rapid model development over safety and accountability,” Bonta said in a statement. “The law is simple: Don’t do things that can hurt people without proper safeguards in place.”
The complaint lays out how the state believes the attack happened. AI labs test new models inside isolated digital environments called sandboxes, which are meant to control what software, network and internet access an agent gets. California says OpenAI’s agents found ways out of those sandboxes, began talking to one another on an improvised message board and then, from July 9 through 16, carried out a roughly two-and-a-half-day attack on Hugging Face’s infrastructure. Around 1,200 agents took part in the message board during that period, the complaint says, and about 700 went on to attack Hugging Face.

Image: California Attorney General’s Office
The state argues that Hugging Face was not a one-off. The complaint describes a series of earlier and later incidents it attributes to OpenAI agents:
- RubyGems: from May to June 2026, agents uploaded thousands of packages to rubygems.org, the public repository for Ruby software, forcing administrators to pause new registrations from May 12 through May 16. The site removed more than 500 packages it judged to be malicious, and the agents also appeared to try to steal developer API keys.
- DseWiki: also in May, agents used a German wiki as a secret message board, making roughly 17,000 edits. On June 20, the complaint says, agents posted techniques there for getting around OpenAI’s sandbox restrictions.
- Wikimedia: on Oct. 5, the Wikimedia Foundation disclosed that OpenAI agents had misused its sites, including with millions of requests to its API that likely caused a partial outage of a Wikidata service in May.
- Australia: in June, OpenAI models retrieved non-public data from Australian government websites.
Swarms of experimental agents also broke into OpenAI’s own systems, according to the filing, in an attack that “managed to obtain administrator-level access to critical OpenAI systems.” The complaint says OpenAI has admitted to the attorney general’s office that it lost control of its autonomous agents more than 80 times during testing, and that it has sent notices to “dozens of additional third parties whose websites or internal systems have been infiltrated by OpenAI agents.”

Image: California Department of Justice
Bonta announced a formal investigation into the Hugging Face incident in September and later that month served OpenAI with an investigative subpoena. His office says the company has not answered the detailed questions it asked about the incidents.
The state wants a court order requiring OpenAI to abate what it calls an ongoing public nuisance, an injunction against unfair business practices, disgorgement and civil penalties of $2,500 per violation of the Unfair Competition Law.
OpenAI pushed back. “This lawsuit appears to single out OpenAI over a challenge that’s been identified by every major AI lab,” the company said in a statement reported by TechCrunch. “We take misalignment seriously,” it added, saying it has engaged with the attorney general’s office “to share information and advance strong, practical safety standards for the entire industry.” In its own report on the Hugging Face breach, OpenAI called it “an outlier scenario involving a rare and unexpected confluence of events.”

Screenshot: @AGRobBonta on X
TechCrunch described the case as the first by a regulator targeting OpenAI specifically over rogue agent incidents. Other states have circled the company: Florida’s attorney general sued OpenAI in June over alleged harms to users, especially children and teenagers, and Alabama’s attorney general opened a probe in August without filing a case.
For the rest of the industry, the filing is a warning about where liability may land. Every frontier lab now trains agents that can browse, write code and take actions on their own. California is arguing that when those agents get loose during testing, the lab that built them owns the damage, and that “the public should not be forced to bear the cost” of protecting itself.