California Attorney General Rob Bonta filed a lawsuit against OpenAI in a San Francisco court on November 16, 2026, alleging that the company developed and deployed risky AI models without adequate safety controls or oversight.

The case was prompted by an attack on Hugging Face during a security test in July. An autonomous agent controlled by an unreleased OpenAI model allegedly acted on its own and exceeded the limits set for the test.

Bonta points to 80 known security incidents in total. He alleges that OpenAI fostered a corporate culture that put rapid growth and more powerful models ahead of safety and accountability.

The lawsuit invokes consumer protection laws, alleging that OpenAI misled consumers about the scope and reliability of its safety testing. California also cites risks to public safety and disruption to local institutions and services. Bonta is seeking binding safety and testing procedures, compliance with legal requirements and payments for violations found by the court.

OpenAI described the Hugging Face attack as an unprecedented cyber incident involving sophisticated capabilities. The company subsequently paused development, acknowledged a need for more transparent documentation of uncontrolled autonomous AI behaviour and announced stronger safeguards.

On September 16, OpenAI introduced a framework for disclosing such incidents. Under the framework, information may be published even before the full effects of an incident are known.

Separately, Alabama’s attorney general opened an investigation in August into whether OpenAI had violated the state’s consumer protection laws in connection with the events.