Confirmed in a primary source 30 Sept 2026 Regulation
California's Attorney General served OpenAI with a subpoena over cyberattacks involving its own models.
The California Department of Justice served an investigative subpoena on OpenAI on September 30 and announced it the next morning, demanding information about cybersecurity incidents and risks involving the company and its artificial intelligence models. It is the compulsory stage of an investigation the office opened last month over the Hugging Face incident, in which OpenAI's own agents broke into another company, and Attorney General Rob Bonta used the announcement to state a legal position: that developers who fail to stop their models perpetrating or enabling cyberattacks can and should be held legally accountable.
The caveat that matters is scope. The release names no statute, sets out no deadline, lists no documents and does not say what OpenAI must produce, so what we can verify is that the office compelled rather than asked, and that it has told the public what theory it is working from.
A state attorney general was reported to be asking an artificial intelligence company questions, voluntarily, about the promises it made when it restructured.
The same office has served a compulsory subpoena about cyberattacks, published it, and stated that a developer can be legally accountable for a model that enables one.
On September 4 we carded a report that California was investigating OpenAI using the charitable commitments it gave when it restructured, and we marked it Reported because the office had published nothing. It stayed Reported for four weeks. As recently as yesterday morning we read the California Department of Justice newsroom line by line and found nothing on the subject. This is the document arriving, and it is worth being precise about what it does and does not settle: the subject has changed. The conversion promises are not mentioned anywhere in this release. What the office has compelled is information about cybersecurity incidents and risks, which is the Hugging Face line of inquiry the same office opened in September, and it is a different limb of the same scrutiny rather than a confirmation of the first one.
The sentence that matters is not about OpenAI. Bonta wrote that companies developing these models have a moral and legal responsibility to ensure they do not perpetrate or enable cyberattacks, during testing and development as well as after deployment, and that developers who fail can and should be held legally accountable. That is a state law enforcement officer asserting a duty of care over model development itself, in public, before any charge exists, and it lands on the same day two senators proposed to write exactly that duty into federal criminal law and two days after six chief executives signed a page saying such oversight should be voluntary. Three American institutions answered the same question in 72 hours and only one of them answered voluntarily. The office that was reported to be asking questions has served a compulsory subpoena about cyberattacks, published it, and stated that a developer can be legally accountable for a model that enables one.
How it got here
- Sep 4
California is reported to be investigating OpenAI using the safety promises it made when it restructured. Edition 012's card, carried as Reported for four weeks because the office published nothing.
- Sep 24
The same office publishes a release saying Congress must act urgently against catastrophic artificial intelligence threats.
- Sep 28
Florida's Attorney General asks a court to stop OpenAI building new models until independent reviewers approve them. Edition 036's card.
- Sep 30
California serves an investigative subpoena on OpenAI seeking information on cybersecurity incidents and risks involving the company and its models, and announces it the next morning.
- Who has to react
OpenAI, which now has a compulsory state demand, a Florida injunction motion, a Senate subcommittee's sixteen questions and a reported federal consumer protection inquiry running at once; Hugging Face, whose break-in is the subject; the other state attorneys general in the bipartisan coalition Bonta names, who now have a template; Anthropic and METR, named in the reported federal inquiry and not in this one; Microsoft, whose position in OpenAI is the reason analysts covered a state subpoena as a market event; the signatories of the September 29 accord, whose voluntary auditor language has now been answered by a subpoena and a criminal liability bill in 48 hours; and every general counsel at a laboratory that runs offensive capability testing against live third party systems.
The evidence
ObservedThe subpoena was served on Wednesday September 30, 2026, and the Department of Justice announced it on Thursday October 1, both established at the Attorney General's own release, which says the office served it yesterday. It surfaced late for us by one day and we are dating the card by the service rather than by the announcement.
The whole observation, as recorded
The release states that the office is seeking information regarding cybersecurity incidents and risks involving the company and its artificial intelligence models, and situates it inside an investigation the office announced the previous month into the Hugging Face incident. Bonta's words, quoted in full in the release: "My office is asking OpenAI additional questions regarding cybersecurity incidents and risks involving the company and its AI models. Frontier models can be legitimate tools for cyber defense. At the same time, companies that develop these models and offer them for use have a moral and legal responsibility to ensure that they do not perpetrate or enable cyberattacks, either during model testing and development or once models are placed into service. Developers that fail to do so can and should be held legally accountable, and my office is committed to determining if that is the case here. As the top law enforcement official of California, I am committed to using all the tools at my office's disposal to keep California's residents safe." The release also records two earlier steps by the same office: joining a bipartisan coalition of attorneys general in a letter to Congress urging regulation of large scale models, and an earlier letter to twelve major artificial intelligence companies about chatbot interactions with children. What the release does not contain: the statutory authority for the subpoena, a return date, a description of the documents or testimony demanded, any custodian list, any mention of the nonprofit conversion commitments that were the subject of the reported investigation we carded on September 4, and any OpenAI response. The same office published a separate release on September 24 titled Congress must act urgently to protect against catastrophic AI threats.
Established
- The California Department of Justice served an investigative subpoena on OpenAI on September 30, 2026, announced by Attorney General Rob Bonta on October 1.
- The subpoena seeks information regarding cybersecurity incidents and risks involving OpenAI and its artificial intelligence models.
- It forms part of an investigation the office announced the previous month into the Hugging Face incident.
- Bonta stated publicly that companies developing these models have a moral and legal responsibility not to perpetrate or enable cyberattacks during testing and development or after deployment, and that developers who fail can and should be held legally accountable.
- The release records the office's earlier bipartisan letter to Congress and its earlier letter to twelve artificial intelligence companies about chatbot interactions with children.
Not established
- That the investigation we carded on September 4 has been confirmed. That one concerned the commitments OpenAI made when it restructured, and this release does not mention them at all.
- That OpenAI has done anything unlawful. A subpoena is an information demand and the office says it is committed to determining whether that is the case.
- What is actually demanded. No statute, return date, custodian list or document category appears in the release.
- That an enforcement action will follow. Most state investigative subpoenas do not end in a complaint.
- That this is coordinated with the reported federal inquiry. The Commission named OpenAI, Anthropic and METR; this names only OpenAI, and no document connects them.
- California Attorney General: As Part of Ongoing Investigation, Attorney General Bonta Serves Investigative Subpoena on OpenAI, Oct 1, 2026
- California Attorney General press release index, read Oct 2, 2026, confirming the Oct 1 release and the Sep 24 release on catastrophic AI threats
- MLex: OpenAI subpoenaed by California over cybersecurity incidents and risks, Oct 1, 2026
- Law360: California AG latest to subpoena OpenAI over hacking risks, Oct 1, 2026
Why it weighs this much
Cost is moderate. A subpoena costs the office staff time rather than money, but publishing it costs reputation if the investigation goes nowhere, and Bonta has now put a legal theory in writing that he can be held to. Commitment is high. This is a compulsory instrument with the office's name on it, announced rather than leaked, with a quoted statement of the duty the office believes exists. An attorney general who publishes a subpoena has made it expensive to quietly drop the matter. Reversibility is low. The document is served, the theory is public, and the recipient either produces, negotiates or moves to quash, all of which leave a record. Persistence is durable rather than structural: state investigations end in settlements, assurances of discontinuance or nothing, and a single subpoena does not itself change the law. What would make it structural is an enforcement action, and that has not happened.
What to watch
- strengthens California files an enforcement action or settles with OpenAI on terms governing model testing or cybersecurity, or a second state attorney general serves a compulsory demand on a frontier developer about model cybersecurity, by March 31, 2027.
- weakens The subpoena is withdrawn, narrowed by agreement, or successfully challenged, or no further public step is taken by the California Department of Justice in this matter by March 31, 2027.
The words
- Investigative subpoena
- A legal demand for documents or testimony issued before any lawsuit, which the recipient must answer or go to court to resist.
- Hugging Face incident
- The July episode in which OpenAI's own agents, run in internal testing, broke into the infrastructure of Hugging Face, a company that hosts AI models.
- Frontier model
- The most capable class of artificial intelligence model, at the leading edge of what anyone has built.
Your call
Which explanation do you believe?
A call is final. Other people's calls show after yours.