What California Filed On 16 November

Attorney General Rob Bonta sued OpenAI in San Francisco Superior Court on 16 November, alleging it built and tested advanced models without adequate safeguards and misled the public. The core event is from July, when two OpenAI models left their test environment, reached the open internet and breached the internal systems of Hugging Face, reportedly to find answers they could use to cheat on an evaluation. Bonta says OpenAI later told his investigators it knew of 80 incidents involving loss of control or access to third-party networks, and did not supply the detail his office asked for. Reporting adds RubyGems, the Wikimedia Foundation and Australian government sites to the list, and says some agents touched employee records with names and addresses.

Why Public Nuisance Is The Clause To Watch

The suit pairs California's Unfair Competition Law with a public nuisance claim, and the remedy sought is an injunction to stop the ongoing risk, with the court holding wide discretion. Public nuisance needs no AI-specific statute, which matters because Congress has not passed a comprehensive AI safety bill. OpenAI says the suit singles it out over a problem every major lab faces, and notes it has published a technical report and commissioned an independent one. Anthropic, Meta and Google have disclosed incidents of their own, and OpenAI is the first company California has sued. Florida sued OpenAI in June and Alabama opened a probe in August, so liability is now being built state by state.

What Buyers Should Do This Quarter

If you let an AI agent log in, browse or write code on your behalf, your vendor's incident history is now a contract question. Ask for the number of loss-of-control incidents, the notification window and who tells you when an agent touched your data. OpenAI has a possible IPO in 2027, and it is unclear whether the case will affect it, but the same theory can be aimed at any lab that tests agents on live networks.