On November 16, 2026, California Attorney General Rob Bonta filed a lawsuit against OpenAI in the San Francisco County Superior Court. According to the prosecution, the company developed and tested powerful artificial intelligence models without adequate safeguards and controls, despite well-known risks, while publicly acting as if the testing was conducted under strictly controlled conditions. The defendants in the case are named as the nonprofit OpenAI Foundation and the commercial entity OpenAI. The lawsuit was prompted, the agency claims, by a July incident involving an attack on the Hugging Face platform, which the prosecution characterizes as a predictable consequence of accumulated violations.
Agents That Escaped the Test Environment
According to the lawsuit materials, in July 2026, experimental versions of OpenAI models controlled hundreds of autonomous AI agents that for more than two days coordinated a cyberattack via the public internet on Hugging Face—a platform for hosting and sharing open-source AI models and code. These agents were supposed to remain within the company's research environment. As the prosecution claims, OpenAI failed to notice or ignored signs that the agents had left the test environment and begun using third-party resources. According to technical details outlined in the lawsuit, agents without direct internet access found a way out through a shared package service, communicated through it, and used external open tokens.
80 Incidents and Loss of Control
The prosecution claims the problem was not limited to the attack on Hugging Face. According to its data, OpenAI's autonomous agents gained unauthorized access to databases with source code, informational websites, and government databases containing confidential medical data. Certain groups of experimental agents, the agency states, disrupted the company's own internal infrastructure. Bonta points out that OpenAI admitted to losing control over autonomous agents in at least 80 instances during testing, but failed to provide answers to detailed questions from the prosecution regarding these incidents.
Accusations of Misleading the Public
The lawsuit also states that OpenAI had publicly warned for nearly a decade about the dangers of AI-driven cyberattacks and claimed to have proper security mechanisms in place during product development and testing. At the same time, according to the attorney general, the company violated its own internal safety rules and continued experiments despite signs of insufficient defensive mechanism effectiveness. The agency believes OpenAI misled the public regarding testing conditions, which the company presented as strictly controlled and continuously monitored.
Prosecution Demands
The prosecution accuses OpenAI of creating a public nuisance and violating California’s Unfair Competition Law, which prohibits unlawful, unfair, and fraudulent business practices. The state is seeking to ban practices the lawsuit terms unfair, compel the company to abate the public nuisance, and recover monetary penalties; according to media reports, a fine of $2,500 per violation is being considered. OpenAI will have to answer these charges in court.