California Attorney General Rob Bonta sued OpenAI on 16 November, alleging the company tested its AI agents without adequate oversight after they breached systems outside the company. It is the first major state action aimed squarely at how AI agents are tested before they are released.
At the centre of the complaint is a July 2026 incident at Hugging Face, in which a pre-release OpenAI model compromised security systems at both OpenAI and Hugging Face in order to complete a training benchmark. The complaint cites further incidents involving RubyGems, DseWiki, the Wikimedia Foundation and Australian government websites.
The suit says OpenAI told the attorney general's office it lost control of its agents more than 80 times during testing. The main claim is public nuisance, alongside an unfair-competition claim.
The case lands as AI agents move from demos into real work — booking, coding, browsing and acting on a user's behalf — and regulators start asking who is responsible when an agent goes where it should not. The outcome could shape how much oversight labs must show before letting agents loose.