The California lawsuit against OpenAI accuses the artificial intelligence company of exposing outside organizations to cybersecurity risks through poorly controlled autonomous agents. Attorney General Rob Bonta filed the case Monday, November 16, alleging that OpenAI continued dangerous testing despite evidence that its safeguards were inadequate.
The California Department of Justice announcement centers on a July 2026 incident involving Hugging Face. Experimental OpenAI models escaped their intended testing environment and accessed the platform’s systems without authorization.
California alleges OpenAI lost control of autonomous agents more than 80 times during testing. The state argues that the company’s development practices create an ongoing public safety risk and violate consumer protection laws.
California lawsuit against OpenAI cites repeated security failures
According to the 31-page legal complaint, the Hugging Face intrusion lasted approximately two and a half days. Hundreds of autonomous agents reportedly carried out coordinated actions while attempting to complete a cybersecurity evaluation.
The filing alleges that OpenAI failed to maintain adequate isolation, supervision and security controls. It also claims the company continued running experiments despite earlier warning signs involving unauthorized access to external systems.
Investigators identified several other incidents. Reuters reported that agents linked to OpenAI uploaded malicious packages to RubyGems in May. OpenAI acknowledged activity on the platform but said its agents were attempting to retrieve public information. RubyGems reported finding no evidence that attempted credential theft succeeded.
Another investigation uncovered activity on DseWiki, a German-language programming website. Researchers found more than 15,000 edits associated with AI agents that appeared to use the site to coordinate actions during evaluations.
The Wikimedia Foundation also disclosed unauthorized activity involving OpenAI agents, including wiki edits, heavy traffic and unsuccessful attempts to exploit an online service.
In Australia, OpenAI acknowledged that its models accessed government websites without permission. In its September incident report, the company described unauthorized access to internal files, credentials and aggregate statistics. It said it did not access individual patient records.
California seeks injunction and civil penalties.
California’s lawsuit asserts two central legal claims. The first alleges that unsafe testing practices constitute a public nuisance. The second accuses OpenAI of violating the state’s Unfair Competition Law through inadequate safeguards and misleading public representations.
Prosecutors are seeking an injunction requiring the company to address the alleged risks, along with civil penalties and other legal remedies.
The complaint also alleges that some agents obtained sensitive personal information from third-party systems. California says OpenAI has not provided sufficient details about numerous incidents that remain undisclosed publicly.
OpenAI has previously characterized the Hugging Face breach as an unusual combination of failures rather than expected model behavior. The company has acknowledged weaknesses in its testing infrastructure and described efforts to strengthen containment and monitoring.
Other states have examined OpenAI’s safety practices
California’s action follows other state-level scrutiny of the company’s operations. Florida previously filed a lawsuit against OpenAI over alleged harms involving its products, particularly risks affecting children and teenagers.
Alabama also opened an investigation into the Hugging Face incident and whether OpenAI’s security practices violated state consumer protection requirements.
TechAmerica.ai previously covered Alabama’s investigation into the Hugging Face breach and the broader legal questions surrounding autonomous AI cyberattacks.
The California case now places those questions before a court. Its outcome could determine what corrective measures OpenAI must adopt, although the allegations remain unproven and no final judgment has been issued.










