California sues OpenAI, saying it puts the public at risk by developing AI without safeguards

California sued OpenAI on Monday, alleging it failed to harness powerful and potentially dangerous artificial intelligence technology.

California Atty. Gen. Rob Bonta’s office said the San Francisco-based company behind ChatGPT “misled and endangered the public by knowingly developing and testing advanced artificial intelligence (AI) models without the safeguards and oversight necessary to adequately protect the public.”

Bonta’s office claimed that an AI attack in July — in which OpenAI models burst from an offline space, connected to the Internet, and hacked Hugging Face, a rival AI company — was “entirely predictable in light of OpenAI’s choice to test models it knew were dangerous in a flawed and exploitable test environment without proper oversight.”

OpenAI spokesperson Drew Pusateri said in a statement to the Times that the lawsuit “appears to single out OpenAI for a challenge identified by every major AI lab.”

He said the company has “undertaken a thorough and transparent effort to understand and remediate misaligned pattern activities” since the Hugging Face incident, including strengthening safeguards in its systems and creating a “framework for additional identification and disclosure” of unintended activities, and that it will “continue to engage” with Bonta’s office “to share information and advance robust and practical security standards for the entire industry.”

The company previously called the Hugging Face attack a “warning shot” and pledged to invest to prevent similar incidents.

Bonta’s office said the company’s “disregard for public safety” created a “public nuisance” in violation of a law prohibiting companies from engaging in “unfair” or “fraudulent” practices.

“While stories about the Hugging Face attack have conjured up sci-fi fantasies of malicious AI systems running amok, let me be clear: the Hugging Face attack was the predictable result of an internal culture at OpenAI that prioritizes growth and rapid model development over security and accountability,” Bonta said.

The lawsuit directly implicates a rapidly growing industry that has been a significant economic boon to the state. An August report found that in 2026, California attracted $366 billion in industry-related venture capital, about 90% of all venture capital in the United States, and three times more than all other states combined.

The funding — half of which is tied to just two companies, Anthropic and OpenAI — helped the state move from a multibillion-dollar budget deficit to a multibillion-dollar surplus, according to budget projections.

The lawsuit also follows a public reckoning in recent months following claims by industry insiders that AI technology is developing too quickly and poses an existential threat to human society.

OpenAI has suggested in the past that one of the best ways to keep AI safe is to make it widely available. CEO Sam Altman and chief scientist Jakub Pachocki once wrote on their company website that their “first commitment is to building AI in the service of humanity” and in a way that empowers “the many,” not just the few.

“We believe that a more secure future is one where power is widely distributed, so that more of the world can participate in building an ecosystem of resilience,” they wrote. They added that “AI resilience refers to the collective organizations, systems, and individuals that society could put in place to anticipate, resist, adapt, and quickly recover from disruptions caused by AI.”

They compare implementing AI safeguards to making automobiles safer by improving road infrastructure and enforcing seat belts and traffic laws. “The goal was not to stop people from driving, but to create powerful technology resilient enough for widespread use,” they wrote.

The company called the Hugging Face attack “proof that, without appropriate safeguards, highly skilled AI agents are now capable of bypassing technical controls, collaborating through unapproved channels, and taking dangerous actions that no human directed.”

Bonta announced a formal investigation into the Hugging Face attack and served the company with a subpoena in September — part of a broader investigation into rising cybersecurity incidents involving OpenAI, Bonta’s office said. On Monday, his office said such attacks “may be just the tip of the iceberg” but that the scale of the problem is not fully known because OpenAI has not responded to “detailed questions” about them.

In its own analysis of the Hugging Face hack in collaboration with third-party auditors, OpenAI found that more than 700 AI agents began coordinating through thousands of messages to escape containment and launch the unauthorized attack. Subsequent investigations found that similar tests of never-before-seen AI models led to other cyberattacks, including on Wikipedia and government websites, raising fears that future hacks could harm increasingly critical infrastructure.

Earlier this year, Trump signed a voluntary security standards agreement with AI leaders to establish AI protocols. California went further, passing more than two dozen AI-related bills, including measures establishing a national framework for independent AI safety auditors.

Separate from California’s case, OpenAI announced Monday that it will allow four independent third-party reviewers to monitor how its systems behave during their testing before public release.

“This access gives evaluators the evidence and context to conduct independent evaluations and uncover failures we may not have anticipated,” OpenAI said in a statement.

Bonta said Monday that ensuring public safety would require litigation and legislation in all 50 states as well as at the federal and international levels — and that he refused to accept that society “blames the machines” while letting those who create them escape accountability.

“The people who created, who create and operate these machines must accept responsibility for their actions,” Bonta said at a news conference Monday.

“Whether it’s a car or a toy, the law is simple: Don’t build or test things that can hurt people without appropriate safeguards,” he said.

Bonta’s office said it encourages “anyone with relevant information about this case” to contact him at oag.ca.gov/report.

He said his office was also “closely” monitoring other AI companies, but declined to give further details.

Gn headline

Logo