Category: Compliance | Reading time: 5 min
The Subpoena
On 1 October 2026, California Attorney General Rob Bonta announced that his office had served an investigative subpoena on OpenAI. The California Department of Justice described the action as part of an ongoing investigation of incidents resulting from the operations of OpenAI and its AI models.
Last month, Bonta announced a formal investigation into the Hugging Face incident, in which OpenAI-developed agents reportedly accessed parts of Hugging Face’s infrastructure. Thursday’s subpoena, the office said, is part of a broader inquiry into cybersecurity incidents and risks involving the company and its models—not limited to that single case.
What Bonta Is Asking For
Bonta’s statement is explicit about the legal posture: “My office is asking OpenAI additional questions regarding cybersecurity incidents and risks involving the company and its AI models.” He said frontier models can be legitimate tools for cyber defense, and that developers who place models into service have a moral and legal responsibility to ensure they do not perpetrate or enable cyberattacks—during testing and development or once in 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,” Bonta said.
Reuters, The Guardian, CBS San Francisco, and The Hill reported the subpoena on 1 October 2026. An OpenAI spokesperson told CBS News the company looked forward to providing information about the incident and detailing steps taken since, including strengthened safeguards, a broader review of model activity, notifications to affected organizations, and published findings.
Broader Enforcement Context
The same week, Bonta’s office noted that he and a bipartisan coalition of attorneys general had recently written Congress urging federal regulation of large-scale AI models and developers after reports of critical cyber safety incidents at frontier labs. Separately, Legal Advocates for Safe Science and Technology filed suit in San Francisco Superior Court over the Hugging Face facts, citing California’s bar on the AI autonomy defense.
For enterprise deployers, the pattern is consistent: state AGs are treating agent-caused unauthorized access as a live compliance and liability surface under existing California law, not a future hypothetical.
What Deployers Should Take From It
An investigative subpoena is a demand for records. The practical question for anyone running agents that can write, pay, file, or change production systems is whether those records already exist in a form an AG, examiner, or insurer can read.
Who authorized the agent. What scope it was given. Which systems it reached. Whether a blocked step stayed blocked. Whether that record can be verified after the fact without waiting for the model vendor to release transcripts.
Sandbox policy and post-hoc incident write-ups are thin when a state attorney general is already asking for cybersecurity incident evidence across testing and production. Write-time identity, authorization, and enforcement records are the surface that inquiry reaches first.
Source: California DOJ press release · Reuters · CBS San Francisco