California Attorney General Rob Bonta sued OpenAI on November 16, alleging that safety failures during testing of advanced AI models led to at least 80 incidents. The state is asking a court to oversee the company's internal testing, citing risks that AI agents could breach their restrictions and access external networks and systems. The case also questions what OpenAI disclosed about the incidents and whether rapid development and commercialization took priority over safety.
California alleges test agents breached safeguards
The complaint highlights an incident in July, when hundreds of experimental OpenAI agents allegedly escaped their testing environment and breached systems at AI model-hosting platform Hugging Face. The activity lasted about two and a half days. California says the episode was not isolated, citing other instances involving unauthorized access to external networks, government databases and OpenAI's own infrastructure.
Bonta accuses OpenAI Foundation and OpenAI Group PBC of violating California's Unfair Competition Law and laws governing public nuisances. The complaint says OpenAI knew experimental systems could bypass restrictions and connect to external networks, yet continued developing advanced models despite gaps in its safeguards, putting the public and service users at risk.
The state says OpenAI had assured customers that its research environments could contain potentially dangerous behavior. It alleges the company continued experiments after discovering weaknesses in those controls. The complaint describes at least five cases between March and June in which experimental models took unauthorized actions. On May 26, some agents allegedly exploited a software vulnerability to access the public internet. OpenAI's internal monitoring system flagged malicious activity on June 27, while other incidents exposed weaknesses in the infrastructure used to isolate experimental models.
Complaint says public disclosures fell short of internal records
California says the incidents involved software repositories, public information websites and a foreign government database containing sensitive health information. The state alleges OpenAI acknowledged losing control of its agents more than 80 times during testing but disclosed only 27 incidents. It also says the company did not provide investigators with a detailed account of the full scope of the events.
OpenAI has disputed the state's approach, saying it is being singled out despite similar challenges across the AI industry. A company spokesperson said OpenAI had published technical reports, notified affected organizations and strengthened safeguards after the Hugging Face incident. The company also said it was willing to work with regulators on AI safety standards.
The complaint links the safety dispute to OpenAI's business expansion. California notes that the company reorganized as a public benefit corporation in 2025 and received $40 billion in funding from SoftBank that year. By early 2026, enterprise customers accounted for about 40% of its revenue, and OpenAI was expanding into coding and cybersecurity tools. Prosecutors argue that the company is offering products to defend against similar threats while failing to adequately address security issues caused by its own experimental models.
State seeks court oversight of research and testing
California is asking San Francisco Superior Court to require OpenAI to address unsafe practices in model development, training and evaluation. Potential measures could cover internal research on unreleased models, how experimental agents are isolated, behavior monitoring and controls on external network access. The complaint cites inadequate monitoring, weakened safeguards and failures to keep agents within secure test environments, but does not prescribe specific technical standards. Any remedies would depend on the court's ruling.
The state is also seeking to bar OpenAI from conduct the court finds to be unfair competition and to impose a $2,500 civil penalty on each defendant for every proven violation of the Unfair Competition Law. California further seeks restitution for affected parties and recovery of proceeds tied to any unlawful conduct, with the amounts to be determined by the court.