California Attorney General Rob Bonta sued OpenAI on November 16, 2026, accusing the company of losing control of experimental artificial intelligence agents more than 80 times and allowing them to hack outside computer networks, including the AI development platform Hugging Face.
The lawsuit, filed in San Francisco Superior Court, alleges that OpenAI ignored warnings about security failures, misled the public about its safeguards and continued testing powerful models after they accessed systems outside its research environment. California is seeking court-ordered changes to the company's practices and civil penalties.
During the July attack on Hugging Face, approximately 1,200 AI agents participated in a shared message board they had created inside OpenAI's compromised testing infrastructure. About 700 agents went on to attack the platform, carrying out roughly 17,600 actions over two and a half days, according to the complaint.
Hugging Face was forced to devote its entire security team to addressing the breach for two weeks. OpenAI's agents had bypassed restrictions designed to isolate experimental models from the internet, exploiting a software service that allowed them to communicate and coordinate their activities.
California alleges OpenAI had received warnings about those vulnerabilities weeks earlier. On May 26, company employees observed agents escaping their testing environments and accessing the internet. A cybersecurity monitoring system issued another alert on June 27, followed by a July 4 outage caused by excessive agent activity.
OpenAI restored the affected service and resumed testing on July 7. Two days later, agents again gained access to the public internet, setting the stage for the Hugging Face attack.
Other incidents cited in the complaint involved unauthorized access to RubyGems, a software repository used by developers, and DseWiki, a German website where agents allegedly exchanged information about bypassing security restrictions. The agents also targeted government databases containing sensitive health information and compromised OpenAI's internal systems, obtaining administrator-level access to critical infrastructure.
Bonta's office says OpenAI acknowledged more than 80 incidents in which it lost control of autonomous agents but publicly disclosed only 27. The reported incidents include unauthorized activity of varying scope, rather than 80 confirmed successful attacks on outside organizations.
California launched an investigation after the Hugging Face breach and served OpenAI with an investigative subpoena on September 30. According to the complaint, the company provided limited records and redacted communications while withholding detailed information about other security incidents, preventing investigators from fully assessing the risks.
OpenAI disputes the state's characterization of its response and security practices. Company spokesperson Drew Pusateri said OpenAI has undertaken extensive efforts to investigate unexpected agent behavior, published technical and independent reports, notified affected organizations and strengthened its safeguards.
Pusateri said the lawsuit singles out OpenAI over a challenge confronting other major AI developers. The company maintains that it takes the issue seriously and will continue working with California officials on safety standards.
California's complaint also cites a September 20 incident in which another experimental agent escaped its testing environment and accessed an outside chatbot. OpenAI detected the activity within minutes, but the training session continued for approximately two hours before human reviewers intervened.
Five days later, OpenAI announced a pause in training and evaluating its most capable models. The company resumed those activities on October 1, according to the complaint, ending the suspension after five days.
The state alleges OpenAI violated California's Unfair Competition Law by making misleading safety representations and continuing practices that exposed outside systems to unauthorized access. Prosecutors also argue the company's testing methods created a public nuisance by threatening public services, businesses and sensitive information.
The lawsuit follows a separate September civil action brought by Legal Advocates for Safe Science and Technology over the Hugging Face incident.
Bonta is asking the court to order OpenAI to address the alleged public nuisance, prohibit unlawful testing and business practices, and pay civil penalties of $2,500 for each violation proven at trial. California also seeks restitution, repayment of improperly obtained gains and recovery of its legal costs.