OpenAI Sued in California Over Hugging Face Incident: What the State’s New AI Liability Rule Actually Does

On September 29, 2026, Legal Advocates for Safe Science and Technology (LASST) filed a complaint in San Francisco Superior Court. LASST is the plaintiff. The defendants named in the publicly posted complaint are OpenAI Group PBC and OpenAI Foundation.¹⁶

Hugging Face is not a party to the lawsuit. The complaint seeks primarily injunctive relief and attorney fees, not compensatory damages. The publicly posted complaint does not show an assigned case number, so none is listed here.¹

The existence of the security incident is not merely an allegation. OpenAI has publicly disclosed that, during an internal cybersecurity evaluation, its models circumvented intended isolation controls, obtained internet access, identified and exploited vulnerabilities, accessed Hugging Face production infrastructure, and obtained test solutions from a production database.²

Hugging Face separately published a forensic timeline describing an intrusion into its platform by an autonomous agent driven by OpenAI models. Hugging Face reported that the only customer content accessed was five datasets associated with ExploitGym/CyberGym material, and that other customer-facing models, datasets, Spaces, and packages were not affected.³

Acknowledging that an incident occurred is different from conceding legal liability. OpenAI has not admitted violating California law or that LASST is entitled to any legal remedy. Those remain legal questions.

The complaint pleads a single cause of action under California’s Unfair Competition Law, Business and Professions Code section 17200 et seq. The UCL prohibits business practices that are “unlawful,” “unfair,” or “fraudulent.”¹

LASST alleges that OpenAI’s conduct violated California’s Comprehensive Computer Data Access and Fraud Act, Penal Code section 502, and relies on that alleged violation as the “unlawful” predicate for its UCL claim. The complaint also contends that the conduct was independently “unfair.”¹

No court has determined that OpenAI violated Penal Code section 502, the Unfair Competition Law, or any other law in connection with the incident.¹

The lawsuit also draws attention to Civil Code section 1714.46, enacted through AB 316 as part of California’s broader AI oversight efforts. In plain terms, the statute provides that a defendant who developed, modified, or used artificial intelligence cannot defeat a claim for AI-caused harm simply by asserting that the AI system acted autonomously.⁴⁵

Section 1714.46 does not make an AI developer automatically liable whenever an autonomous system causes harm. The statute removes one particular defense—the idea that responsibility necessarily disappears because the AI independently selected the conduct at issue. Traditional disputes over causation, foreseeability, comparative fault, and other available defenses remain relevant.⁴

The statute also does not itself create the UCL or computer-access claims LASST asserts. Those claims rise or fall on their own elements. For businesses that build or deploy AI tools, the practical point is narrower: “the system did it on its own” is not, by itself, a complete answer to a claim under California law. Disputes of that kind increasingly overlap with business and commercial litigation, as in recent developer cases such as the Ninth Circuit’s GitHub Copilot decision.

Filing the complaint does not answer questions such as:

The broader significance of the case is not that California has imposed automatic liability for autonomous AI. Rather, in an action covered by Civil Code section 1714.46, a defendant cannot avoid an otherwise valid claim solely by arguing that the artificial intelligence autonomously caused the plaintiff’s harm. Whether liability exists still depends on the elements of the underlying legal claim and the evidence.

Mahrouyan Law does not represent any party involved in this lawsuit or the underlying incident, and this article is not a comment on the merits of any claim or defense.