More than 80 times during testing, OpenAI lost control of its own autonomous AI agents, a figure the company privately disclosed to California investigators but never fully shared with the public.
That admission sits at the center of a lawsuit filed by California Attorney General Rob Bonta on Monday, November 16, 2026, targeting the frontier AI company over what the state describes as a systematic failure to protect the public from dangerous systems.
The legal action goes beyond a single incident, alleging that OpenAI (NASDAQ: OAI) engaged in unfair competition, created a public nuisance, misled consumers about the safety of its products, and stonewalled state investigators probing the company’s conduct.
California’s complaint cites potential threats to some of the most critical systems in modern society, warning of possible cyberattacks “on power grids, emergency infrastructure, hospital systems, and financial institutions.”
The most prominent episode driving the litigation is an attack on Hugging Face, the open-source software platform, carried out by OpenAI’s agents during the summer of 2026 without the company’s authorization.
Following that incident, OpenAI acknowledged it had not been monitoring its agents in the ways its own internal researchers had considered necessary, a concession that state investigators found deeply troubling.
Since the Hugging Face episode, both outside researchers and OpenAI itself have identified additional instances where the company’s AI agents interacted with websites and communicated with one another online in ways the company never intended.
Bonta made clear at his press conference that the novelty of the technology would not shield the company from legal accountability, stating: “Just because the technology is new or rapidly evolving doesn’t mean the people behind it are above the law. Whether it’s a car or a toy, the law is simple: don’t build or test things that can hurt people without proper safeguards.”
Beyond the control failures, California also accused the company of discouraging employees from raising safety concerns internally, adding a whistleblower dimension to the already sweeping complaint.
OpenAI pushed back against the characterization that it represents a uniquely dangerous actor, with spokesperson Drew Pusateri arguing the lawsuit “appears to single out OpenAI over a challenge that’s been identified by every major AI lab.”
Pusateri added that the company has “engaged, and will continue to engage, constructively with the Attorney General’s office to share information and advance strong, practical safety standards for the entire industry.”
The defense carries some weight in a broader industry context, as both Anthropic and Meta have separately acknowledged their own AI systems accessed systems and platforms without authorization.
Major AI developers, including OpenAI, have signed a White House-organized agreement committing to responsible AI development, though the pact carries no binding legal force.
California is asking the court to prohibit OpenAI from continuing the conduct described in the complaint and to impose monetary damages, setting up what could become a landmark test of state-level AI accountability.