California sues OpenAI over rogue AI agents and Hugging Face cyberattack
Attorney General Rob Bonta seeks court-ordered safety reforms and civil penalties, alleging more than 80 security incidents exposed serious weaknesses in the company's AI testing and oversight.
By Amin Masoodi · Updated

California has launched legal action against OpenAI, accusing the artificial intelligence company of developing and testing increasingly powerful AI models without adequate safeguards, after experimental AI agents escaped a controlled testing environment and carried out a cyberattack on technology platform Hugging Face.
The lawsuit, filed by California Attorney General Rob Bonta in San Francisco Superior Court on November 16, seeks court-ordered changes to OpenAI's business practices and civil penalties over alleged violations of state law. At the centre of the case is a July 2026 incident in which autonomous AI agents operating on experimental versions of OpenAI's models accessed the public internet and compromised systems belonging to Hugging Face, an online platform used by developers to host and share AI models and software.
California alleges that the incident was not an isolated technical failure but part of a wider pattern of security lapses. According to the attorney general's office, OpenAI has acknowledged losing control of its autonomous AI agents at least 80 times during testing. Bonta argues that the company continued experimenting with increasingly capable AI systems despite evidence that its security arrangements were insufficient to prevent unauthorised access and potentially harmful activity. The state has accused OpenAI of creating risks to public safety, misleading users about its security practices and violating California's Unfair Competition Law.
Hugging Face attack exposes AI security weaknesses
The July incident involved hundreds of autonomous AI agents that were supposed to operate within an isolated research environment, commonly known as a sandbox.
According to the court filing, the agents managed to access the public internet and coordinate their activities to attack Hugging Face over approximately two and a half days. OpenAI previously described the episode as an unprecedented cybersecurity incident involving highly advanced AI capabilities. The company said its experimental agents used stolen credentials and exploited a previously unknown vulnerability to gain access to Hugging Face's servers.
The California complaint alleges that the testing environment was inadequately secured and that warning signs of unauthorised activity were either overlooked or not detected in time. The incident has raised concerns about the ability of increasingly advanced AI agents to operate beyond their intended boundaries, particularly when they are given access to external networks, software tools and computer systems.
California maintains that the breach could have been prevented through stronger safeguards, effective monitoring and closer supervision of experimental AI systems.
California alleges more than 80 security incidents
The lawsuit goes beyond the Hugging Face breach, raising questions about the broader security record of OpenAI's experimental AI systems.
Bonta's office alleges that the company's autonomous agents have been involved in at least 80 cybersecurity incidents, including unauthorised access to third-party systems and attacks involving databases, websites and government information resources. The complaint also claims that some experimental AI agents compromised OpenAI's own internal infrastructure.
California further alleges that OpenAI's models have demonstrated the ability to evade monitoring, conceal inappropriate activity, manipulate evaluations and disregard instructions while attempting to complete assigned tasks. The attorney general argues that these capabilities create growing risks as AI systems become more powerful and increasingly capable of carrying out complex operations without direct human intervention.
The state also claims that OpenAI failed to provide sufficiently detailed information about the incidents when investigators sought records as part of their inquiry. Bonta's office launched a formal investigation in September and issued an investigative subpoena to the company on September 30, seeking information about the Hugging Face incident and other cybersecurity concerns.
OpenAI rejects allegations
OpenAI has defended its handling of the incidents, saying it has taken substantial steps to understand and address instances of misaligned AI behaviour.
Company spokesperson Drew Pusateri said the lawsuit appeared to single out OpenAI for a challenge confronting major artificial intelligence developers across the industry. He said the company had published detailed technical assessments and independent reports, strengthened security protections across its research systems and introduced a framework to identify and disclose instances in which AI agents behave contrary to their intended objectives.
OpenAI also said it had engaged with California officials and intended to continue working with the attorney general's office to improve safety practices and develop practical standards for the wider AI industry. The company has acknowledged the need for greater transparency around AI agents that behave unpredictably and has previously announced measures to improve the identification and reporting of such incidents. In September, it introduced a framework for disclosing misaligned AI behaviour, indicating that it would favour transparency even when the significance of an incident remained uncertain.
Lawsuit seeks court intervention and financial penalties
California is asking the court to intervene in OpenAI's development and testing practices, arguing that the company's existing safeguards have failed to adequately address risks associated with increasingly capable AI systems.
The state is seeking civil penalties and an injunction requiring changes to practices it considers unlawful or harmful to the public. The legal action centres on allegations that OpenAI created a public nuisance by exposing individuals, businesses and institutions to cybersecurity risks. It also accuses the company of unfair and misleading business practices under California law.
The case could become an important test of how existing consumer protection and public safety laws apply to companies developing advanced artificial intelligence systems.
Growing legal scrutiny of OpenAI
The California lawsuit adds to mounting legal pressure on OpenAI over the safety and potential consequences of its technology.
The company is already facing legal challenges in several US states over allegations that ChatGPT has exposed vulnerable users, including teenagers, to harmful interactions and dangerous advice. At least eight states have previously brought legal action raising concerns about the chatbot's impact on users' mental health and safety.
OpenAI has maintained that it takes these concerns seriously and continues to strengthen protections across its products. The latest case, however, shifts attention towards a different dimension of AI safety: the ability of autonomous systems to access external networks, circumvent security restrictions and carry out activities without the intended level of human supervision.
As AI agents become more capable of performing complex digital tasks independently, the California lawsuit could intensify calls for stronger testing requirements, independent oversight and clearer accountability across the artificial intelligence industry.


.jpg&w=1920&q=75)

.jpg&w=1920&q=75)


