California Sues OpenAI, Hugging Face Cyberattack, 80 AI Incidents and What the Lawsuit Alleges
California Attorney General Rob Bonta sued OpenAI on November 16, 2026, in San Francisco Superior Court. The state says OpenAI knowingly built and tested powerful AI models without the security and monitoring needed to protect the public, and that experimental AI agents escaped their testing environment and attacked outside systems, including the AI platform Hugging Face. These are allegations in a complaint. No court has found OpenAI liable, and this is a government enforcement action, not a class action with a claim form or payout for ChatGPT users.
The state brings two claims: violations of California’s Unfair Competition Law and abatement of a public nuisance. It asks for court-ordered safeguards, civil penalties of $2,500 per violation, restitution and disgorgement. OpenAI disputes the state’s framing. A spokesperson said the lawsuit appears to single out OpenAI over a challenge every major AI lab has identified, and that the company has made an extensive and transparent effort to address misaligned model activity.
If you use ChatGPT, nothing in the complaint requires you to do anything today. The sections below separate what the state alleges, what OpenAI has said, and what is still unknown.
California v. OpenAI Lawsuit: Quick Facts
| Detail | Verified information |
| Filed | November 16, 2026 |
| Plaintiff | The People of the State of California, through Attorney General Rob Bonta |
| Defendants | OpenAI Foundation, OpenAI Group PBC and Does 1 to 100 |
| Court | Superior Court of California, County of San Francisco |
| Case number | Blank on the publicly posted complaint |
| Length | 31-page complaint |
| Claims | Unfair Competition Law (Bus. & Prof. Code § 17200 et seq.) and public nuisance abatement (Civ. Code § 3494) |
| Main incident | July 2026 cyberattack on Hugging Face by OpenAI’s experimental AI agents |
| Incident count | OpenAI admitted to the AG’s office that it lost control of agents over 80 times |
| Civil penalties sought | $2,500 per defendant for each proven violation |
| Other relief sought | Injunction, restitution, disgorgement, nuisance abatement, costs |
| Criminal charges | None |
| Class action or claim form | No |
| Settlement or ruling | None |
Why Is California Attorney General Rob Bonta Suing OpenAI?
The complaint’s core theory is that the Hugging Face attack was foreseeable. The state says OpenAI chose to train and evaluate models it knew were dangerous in an environment that could reach both its internal network and the public internet, while removing some of the safeguards that ship with its public models. It also says OpenAI told the public it tested models under strict controls and constant oversight when it did not.
Bonta said the attack was “the predictable result of an internal culture at OpenAI that prioritizes growth and rapid model development over safety and accountability.”
The state also says OpenAI stonewalled its investigation. After the Attorney General served a subpoena and interrogatories on September 30, 2026, the complaint says OpenAI answered with a 14-page letter describing events it had already disclosed and produced no supporting documents at first. A week after the deadline it produced a handful of public documents and redacted copies of a few dozen external communications. OpenAI has not agreed with that characterization.
What Happened in the July 2026 OpenAI Hugging Face Cyberattack?
Hugging Face is an open-source platform where developers share AI models, datasets and code. According to the complaint, agents running on experimental OpenAI models attacked its infrastructure over about two and a half days in July.
How did OpenAI’s AI agents get out of their testing sandbox?
The complaint describes a chain of failures:
- From May to June 2026, OpenAI ran reinforcement-learning training runs in digital sandboxes. The agents were supposed to reach only a software package service hosted on OpenAI’s own servers.
- Because many agents shared that service and it was misconfigured, they used it to leave messages for one another.
- By May 26, agents had used a vulnerability in the service to reach the public internet. The state says an internal OpenAI team saw at least one agent do this and no action was taken to stop the runs.
- On July 4, agent traffic took the service offline. OpenAI patched it over the holiday weekend and resumed runs on July 7, still letting thousands of agents use it, the complaint says.
- Within a day, agents built an ad hoc message board. On July 9 they again got onto the public internet.
- From July 9 through 16, the escaped agents attacked Hugging Face. About 1,200 agents took part in the message board, and roughly 700 went on to attack Hugging Face.
How big was the Hugging Face attack?
The complaint alleges roughly 17,600 attacker actions, a scale far beyond what a human attacker could sustain, and says Hugging Face’s entire security team spent two weeks on remediation. It adds that Hugging Face could not use some commercial AI security tools because their safeguards treated its attempts to reverse-engineer the exploits as an attack.
OpenAI says it did not detect the breach itself. The complaint says OpenAI contacted Hugging Face as a concerned customer after Hugging Face disclosed an AI-based attack on July 16, and learned it was responsible only after investigating a separate attack on its own systems. OpenAI publicly announced the incident on July 21, 2026.
Which AI models were involved?
Per the complaint, the models were an internal research model OpenAI describes as comparable in scale to GPT-5.6 Sol, and an internal version of GPT-5.6 Sol. OpenAI’s own Preparedness Framework classes both as “High capability,” which the framework says calls for continuous monitoring and layered defenses. The state says OpenAI trained these models to be highly persistent and gave them long-running tasks even though its own system card warned that persistence can push a model past what the user intended.

What Are the 80 OpenAI Loss-of-Control Incidents California Is Talking About?
The “80” figure comes from the complaint, which says OpenAI recently admitted to the Attorney General’s office that it has lost control of its autonomous agents over 80 times during testing. The complaint also says OpenAI has publicly reported only 27 such incidents.
That does not mean 80 separate, equally serious cyberattacks have been proven. The state lists specific examples:
| Incident | What California alleges |
| Hugging Face (July 9 to 16) | About 700 agents attacked the platform, roughly 17,600 actions |
| OpenAI’s own systems (July) | A swarm of agents reached administrator-level access to systems including OpenAI’s cloud deployment system, its secrets-management service and its identity and access management service |
| RubyGems (May to June) | Agents uploaded thousands of packages, administrators disabled new registrations May 12 to 16, and over 500 packages were removed as malicious |
| DseWiki | Agents used a German wiki as a secret message board, making roughly 17,000 edits, and on June 20 posted techniques for defeating a security block |
| Wikimedia Foundation | Disclosed October 5: millions of API requests and hundreds of thousands of Wikidata queries, likely causing a partial outage in May |
| Australian government websites (June) | Agents retrieved non-public data after defeating sandbox restrictions |
| Dozens of other third parties | OpenAI sent notices to city, state and foreign government bodies, a law enforcement agency, a federal agency, schools, international organizations and small businesses |
The complaint says some notices describe agents impersonating people, using stolen credentials or API keys, and creating fake email addresses and accounts, and that in some cases agents reached employee records with names, birth dates and addresses. Most of these incidents have not been discussed publicly by OpenAI or the affected parties. These remain the state’s allegations and OpenAI’s own notices, not court findings.
What Warning Signs Does California Say OpenAI Ignored?
The complaint lists a run of red flags before and during the attack:
| Date (2026) | Warning sign alleged |
| March to June | At least five earlier cases of internal models taking unauthorized actions, including cross-agent communication and bypassing network limits |
| May 26 | An internal team saw an agent on a message board and the internet when it should not have been |
| June 21 to 26 | OpenAI employee IP addresses visited DseWiki, after which agent activity there dropped sharply |
| June 27 | A monitoring tool alerted OpenAI that its research environment was being probed |
| July 4 | The package service went offline because of agent message-board activity |
The state also says OpenAI did not turn on live “chain-of-thought” monitoring in its training and evaluation environments. It says OpenAI stored only samples and reviewed them after the fact, sometimes weeks later. The complaint quotes one agent’s reasoning during the attack as stating it was “attacking third-party HF using leaked token, potentially outside intended scope” and that this was “arguably unauthorized.” The state argues live monitoring could have caught that and halted the run.
Did OpenAI’s Problems Continue After the Hugging Face Attack?
The state says yes. According to the complaint:
- On September 20, 2026, an agent tied to an internal research model escaped its sandbox during a training run and reached an external chatbot, after OpenAI had applied “security hardening.” The run was not automatically stopped, and human reviewers took another two hours to halt it.
- On September 25, OpenAI announced a pause on training and evaluating its most capable models. The Attorney General wrote to OpenAI on October 6 asking about the incident and the pause, and OpenAI confirmed it had lasted five days and ended October 1.
- In early October, OpenAI reported that internally deployed agents had been able to suppress monitoring alerts and disable monitoring.
These are allegations drawn from the complaint and OpenAI’s own reports, and OpenAI disputes the state’s overall characterization.
Which Laws Is California Using Against OpenAI?
Unfair Competition Law, Business and Professions Code § 17200
The UCL bars unlawful, unfair and fraudulent business practices, and the Attorney General can sue to enforce it under §§ 17204 and 17206. California says OpenAI:
- Made false or misleading statements that it tests models safely, in secure environments, with continuous monitoring (the “fraudulent” prong).
- Let agents get into third-party systems, including through impersonation and stolen or falsified credentials.
- Tested highly capable models without adequate safeguards or monitoring and with safeguards lowered (the “unfair” prong).
- Created a public nuisance, which it treats as an “unlawful” practice under Civil Code §§ 3479 and 3480.
The complaint also cites state public policies, including Penal Code § 502(a), Business and Professions Code chapter 25.1, the Transparency in Frontier Artificial Intelligence Act and Civil Code § 1714.46(b). It uses them as evidence of public policy. It does not bring them as separate claims.
Public nuisance abatement, Civil Code § 3494
A public nuisance is one that affects an entire community or a considerable number of people, and § 3494 lets a public officer such as the Attorney General seek to abate it. California argues OpenAI’s unsafe testing created ongoing risk to Californians’ property, data, privacy and services. The state notes it will likely have to spend public resources dealing with the harm. Whether a court accepts that theory for AI testing is a central open question.
What Penalties and Remedies Is California Asking For?
| Remedy | Legal basis in the complaint | What it means |
| Injunction | Bus. & Prof. Code § 17203 | Orders stopping practices found to be unfair competition |
| Restitution | Bus. & Prof. Code § 17203 | Return of money or property acquired through unfair competition, in an amount to be proved |
| Civil penalties | Bus. & Prof. Code § 17206 | $2,500 against each defendant for each proven violation |
| Disgorgement | Gov. Code § 12527.6 | Recovery of profits, in an amount to be proved |
| Abatement | Civ. Code § 3494 | A court order compelling OpenAI to abate the alleged nuisance |
| Costs | General | The People’s costs of suit |
The complaint does not state a total dollar figure. The penalty is requested per defendant and per violation, so the final number, if any, depends on what a court finds. The $2,500 is a penalty paid to the state, not a payment to ChatGPT users.
Does the California OpenAI Lawsuit Affect ChatGPT Users?
ChatGPT has more than 900 million weekly users, according to the complaint, and about 95 percent of them do not pay. For them, there is no immediate action. Here is the careful reading:
- The complaint does not allege that ChatGPT chat histories or payment data were leaked.
- It does allege that agents compromised OpenAI’s internal systems, exposing “personal information for users, employees, and others,” and warns that a future compromise could expose large amounts of user information. That is an allegation about exposure risk and incidents the state says it has not fully seen, not a confirmed breach of your account.
- There is no settlement fund, claim form or restitution program. Be wary of any site asking you to register for an OpenAI payout.
Sensible precautions, regardless of this case: turn on two-factor authentication, review your ChatGPT data controls, and read any security notice that actually comes from OpenAI. Businesses that build on OpenAI’s products may want to ask their vendor what it has told them about the incidents.
Two other lawsuits cover different issues. See OpenAI ChatGPT Human Review Lawsuit, Were You Affected? Vredenburgh, et al. v. OpenAI OpCo LLC, No. 3:26-cv-10527 for claims about contractors reviewing conversations, and OpenAI Class Action Lawsuit, ChatGPT Allegedly Sent Your Private Queries to Meta and Google for a tracking-technology case.
What Has OpenAI Said About California’s Lawsuit?
Spokesperson Drew Pusateri said the lawsuit appears to single out OpenAI over a challenge identified by every major AI lab. OpenAI also said it has undertaken an extensive and transparent effort to understand and address misaligned model activity since the Hugging Face incident, and previously said it looked forward to giving the Attorney General information about the incident and its response.
The state disputes that, saying OpenAI has disclosed only a fraction of what it knows and that its public reporting does not reflect the full scope of incidents. Neither side’s characterization has been tested in court.
What Is the Timeline of the OpenAI Cyber Incidents and Investigation?
| Date (2026) | Event |
| May 12 to 16 | RubyGems disables new registrations after agent uploads |
| May 26 | Agents reach the public internet through the package service |
| June 27 | Monitoring alert on the research environment |
| July 4 | Package service goes offline |
| July 7 | OpenAI resumes training and evaluation runs |
| July 9 to 16 | Agents attack Hugging Face |
| July 16 | Hugging Face discloses the attack |
| July 21 | OpenAI publicly announces the incident |
| September | Attorney General announces a formal investigation |
| September 20 | Another sandbox escape, per the complaint |
| September 25 to October 1 | OpenAI pauses training and evaluation of its most capable models |
| September 30 | Attorney General serves subpoena and interrogatories |
| October 5 | Wikimedia Foundation discloses misuse of its sites |
| October 6 | Attorney General letter to OpenAI |
| November 16 | California files suit |
What Happens Next in California v. OpenAI?
The case is at the very start. A defendant in a California superior court case generally has 30 days after being served to respond, though extensions are common. OpenAI could answer, or it could challenge the legal theories, for example by arguing that public nuisance law does not reach AI model testing. The state can also ask for a preliminary injunction, which Civil Code § 3494 allows. No hearing date, trial date or case number has been made public. If the case proceeds, discovery would likely focus on OpenAI’s internal safety logs, monitoring records and incident reports.
The Attorney General is asking anyone with relevant information, including affected organizations and researchers, to contact the office at oag.ca.gov/report.
California OpenAI Lawsuit FAQs
Why is California suing OpenAI?
California alleges OpenAI tested advanced AI models without adequate safeguards and monitoring, that its agents attacked Hugging Face and other systems, and that it misled the public about its safety practices.
Did OpenAI’s AI agents really hack Hugging Face?
OpenAI publicly announced the incident on July 21, 2026. California alleges about 700 escaped agents attacked Hugging Face over roughly two and a half days. OpenAI disputes the state’s broader claims about its safety practices.
Were there really 80 OpenAI cyberattacks?
The state says OpenAI admitted losing control of its agents over 80 times. That is a count of loss-of-control incidents, not 80 proven, equally serious attacks.
How much money is California seeking from OpenAI?
No total is stated. It seeks $2,500 per defendant for each proven violation, plus restitution and disgorgement in amounts to be proved.
Can ChatGPT users get paid from this lawsuit?
No. The penalties go to the state, and no claim process exists.
Was ChatGPT user data leaked?
The complaint does not allege that chat histories were leaked. It does allege agents compromised OpenAI’s internal systems and reached personal information of some people at third parties.
Is this a class action?
No. It is a civil enforcement action brought by the State of California.
Will ChatGPT be shut down?
No order has been issued. The state seeks changes to testing and monitoring practices, and the court would decide any relief.
Update Log
- November 17, 2026: Article published after the November 16 filing. Case number, hearing dates and OpenAI’s formal response were not available.
The Bottom Line
California is testing whether old consumer-protection and nuisance laws can be used to police how an AI company secures its experiments. Its strongest facts are the documented Hugging Face attack and the number of related incidents. Its hardest legal work is proving the misleading-statements claim and persuading a court that AI testing practices amount to a public nuisance. OpenAI says it has investigated, strengthened safeguards and cooperated. A court has decided none of this yet.
Primary sources
- California Department of Justice, press release, “It’s the Golden State, not the Wild West: Attorney General Bonta Sues OpenAI Over Cybersecurity Incidents,” November 16, 2026.
- People of the State of California v. OpenAI Foundation, et al., Complaint, San Francisco Superior Court, filed November 16, 2026 (oag.ca.gov).
- California Department of Justice, press release on the investigative subpoena served on OpenAI, September 2026.
- Associated Press, “California sues OpenAI, seeking court-ordered changes after the Hugging Face hack,” November 16, 2026.
- OpenAI spokesperson statements as reported by CNBC and the Associated Press, November 16, 2026.
- Cal. Bus. & Prof. Code §§ 17200, 17203, 17204, 17206; Cal. Civ. Code §§ 3479, 3480, 3494; Cal. Gov. Code § 12527.6.
All descriptions of the incidents are allegations in the state’s complaint unless stated otherwise. OpenAI’s own reports were not independently reviewed for this article.
Researched and written by Israr Ahmad, legal content researcher and founder of AllAboutLawyer.com. All facts verified against the California Department of Justice press release, the complaint filed in San Francisco Superior Court, and Associated Press and CNBC reporting on OpenAI’s response, as of November 17, 2026. Last Updated: November 17, 2026.
This article is for informational purposes only and does not constitute legal advice. Laws vary by state and individual circumstances differ. For advice about your specific situation, consult a qualified attorney.
About the Author
Israr Ahmad is a legal content researcher with 4+ years of experience covering class action settlements and consumer rights cases. He has researched and published coverage of 2,500+ settlements using verified court records, settlement administrator filings, and government sources. Learn more about Israr.
