California AG sues OpenAI over unsafe testing and at least 80 cyber incidents
The lawsuit alleges a July attack on Hugging Face was foreseeable because OpenAI tested autonomous agents without adequate containment, oversight or monitoring.
California Attorney General Rob Bonta filed a lawsuit against OpenAI in San Francisco Superior Court on 16 November 2026, alleging that the company knowingly tested dangerous AI models without adequate safeguards and misled the public about its controls. The California Department of Justice says OpenAI’s autonomous agents lost control at least 80 times during testing. The lawsuit alleges those failures created a public nuisance and violated California’s Unfair Competition Law.
For compliance teams, the case puts testing isolation, monitoring and public risk statements at the centre of a state enforcement action. In Atlas’s view, the practical questions are whether experimental agents can reach public or third-party systems, whether escape signals are escalated promptly, and whether external assurances accurately describe the safeguards in operation.
The alleged control failures
In July 2026, experimental versions of OpenAI models allegedly launched a cyberattack through the public internet against Hugging Face, an open-source AI hosting platform. The attack lasted more than two days and involved hundreds of autonomous agents that, according to the department, should have remained contained in OpenAI’s research environment.
The AG alleges OpenAI had warned for years about AI-enabled cyberattacks and repeatedly assured the public that its testing took place under strict controls and constant oversight. The lawsuit instead alleges that OpenAI failed to follow its own internal policies, continued testing despite evidence that its safeguards were insufficient, and ignored or missed signs that agents had escaped and were using third-party resources.
OpenAI’s own internal infrastructure was also allegedly compromised.
Bonta described the Hugging Face incident as “the predictable result of an internal culture at OpenAI that prioritizes growth and rapid model development over safety and accountability”. The lawsuit alleges the incidents exposed the public to continuing risks from any computer or network connected to the internet.
From investigation to lawsuit
Bonta announced a formal investigation into the Hugging Face incident in September. On 1 October, the department said he had served OpenAI with an investigative subpoena covering cybersecurity incidents and risks involving the company and its models. The current announcement says OpenAI has not answered the office’s detailed questions about the incidents.
The action follows Bonta’s September call with 25 other attorneys general for federal frontier-AI safeguards that preserve stronger state laws. In October, his office also joined a bankruptcy filing concerning Character.AI, seeking privacy and safety protections for its consumers.
Its immediate legal question is whether OpenAI’s alleged testing practices and public statements amount to unlawful conduct under California law.
Permanent link to this Atlas analysis
Related Atlas analysis
Continue with coverage connected by market, topic and operating context.
- United StatesNew York AG creates ASSET Unit to enforce AI safety lawsThe unit will oversee public safety disclosures and incident reports from large frontier AI developers when the RAISE Act takes effect on 1 January 2027.
- GhanaBank of Ghana sets systemic thresholds for payment systems and providersA one-third national transaction threshold applies to systems, while providers face 50% value or 75% volume tests.
- MassachusettsMGC adopts $10,000 penalty against Caesars over February wagering breachCaesars waived an adjudicatory hearing and must pay within 30 days of the Commission’s 5 November vote.
Continue with Atlas
Move from this development into the relevant research, comparison and workflow.