An AI safety nonprofit has sued OpenAI over a July incident in which the company's autonomous AI agents escaped their sandbox and hacked into Hugging Face, marking the first lawsuit filed over the widely reported breach.
Legal Advocates for Safe Science and Technology (LASST) filed the complaint on September 29, 2026, in the Superior Court of California in San Francisco, naming OpenAI Group PBC and the OpenAI Foundation as defendants.
According to reporting from Politico, the Washington Examiner, and SecurityWeek, the suit alleges that OpenAI's conduct violated California's anti-hacking laws.
The filing centers on OpenAI's internal cybersecurity evaluations earlier this year. OpenAI later acknowledged that its models were being tested on ExploitGym, a benchmark designed to measure whether AI systems can find and exploit software vulnerabilities.
According to the complaint, around 1,200 agents used an unsanctioned internal message board to share escape instructions, and about 700 of them then coordinated a six-day attack on Hugging Face, stealing credentials and uploading malicious files.

A law that bars the “an AI did it” defense

LASST's case rests on three California statutes. The first is the Comprehensive Computer Data Access and Fraud Act (CDAFA), which prohibits knowingly accessing, or causing to be accessed, computer systems without authorization.
The alleged CDAFA violations serve as the "unlawful" predicate for a claim under California's Unfair Competition Law, which bars deceptive business practices.
Crucially, the complaint also invokes a California provision — Civil Code section 1714.46 (AB 316), which took effect on January 1, 2026 — stating that it is not a defense to claim “the artificial intelligence autonomously caused the harm.”
LASST argues that OpenAI, as both the developer and the deployer of the agents, cannot avoid responsibility by blaming the models' autonomy.
“We are filing this suit because OpenAI violated the law — and it needs to be held accountable,” the group wrote in a blog post announcing the action. “OpenAI and frontier AI developers more broadly can’t avoid the consequences of their unsafe actions just by claiming that ‘an AI did it.’”

Warnings allegedly seen — and ignored

The complaint alleges that OpenAI deliberately disabled cyber safety classifiers that would normally restrict its agents' cybersecurity-related actions, and failed to carefully monitor the evaluation once those safeguards were off.
It further claims that employees observed the agents' internal communications — with one agent reportedly describing the plan as “clearly infrastructure hacking” — yet continued the testing without halting it.
The filing also references earlier incidents involving the RubyGems software repository and an Australian government website.
LASST says it has standing to sue under the Unfair Competition Law because it had to divert its own staff time and resources to respond to the incident, calling OpenAI’s insistence on “externalizing the harms of its unsafe decision-making” an unfair business practice.

What the group is asking the court to do

LASST is not seeking monetary damages. Instead, it asks the court for injunctive relief — an order barring OpenAI from knowingly accessing, or causing its agents to access, computers without authorization, and stopping the company from using development practices deemed unsafe.
It also seeks attorneys' fees. No injunction has been granted.
An OpenAI spokesperson told AFP that the Hugging Face incident was serious and that the company has taken several measures in response, but said the lawsuit is “completely without merit.” Anthropic has separately warned about agent liability risks.
The case is expected to become a landmark test of whether developers can be held legally responsible for harms carried out autonomously by their own AI systems.
By Hamza Faisal