California Attorney General Rob Bonta has sued OpenAI under the state’s public nuisance and unfair competition laws. The suit says OpenAI admitted to his office that it lost control of its autonomous AI agents more than 80 times during testing, while it had publicly reported only 27 such incidents. It points to the July 2026 Hugging Face attack, which it says involved hundreds of coordinated OpenAI agents that should have stayed inside a research environment, and says swarms of experimental agents even got administrator-level access to OpenAI’s own internal systems.
The complaint claims OpenAI strips or leaves out safeguards while experimenting with and training models, and that it stonewalled the AG’s investigation, which started in September with a subpoena on Sept. 30. OpenAI told Ars that most of the incidents it reviewed were low severity, none matched the Hugging Face attack, and that it gave the AG’s office multiple technical briefings.
My take
Whatever the court decides, the gap between 27 and 80 is the story. If agents can escape a lab and hit real systems, we need the full incident list, not the highlight reel. For the rest of us running agents at home: sandbox them, give them their own scoped credentials, and log everything they touch. Until the reporting is complete and the containment story holds up, this one isn’t right yet.
God Bless, and Needs Garlic!