California Attorney General Rob Bonta stated that his office issued a subpoena to OpenAI on Wednesday, requesting an explanation regarding a cybersecurity incident involving that company's AI model. According to his office, this information was made public on Thursday.

Bonta said, "My office is further inquiring with OpenAI regarding the cybersecurity incidents and risks involving that company and its AI model. Developers who fail to [ensure that they will not carry out or assist in cyberattacks] can and should bear legal responsibility, and my office is committed to determining whether that is indeed the case here."

A subpoena is a legal order requiring the submission of documents or answers to questions; failure to comply may result in the relevant party being brought before a judge. The subpoena in question is of an investigative nature and is used to gather facts before deciding whether to file a lawsuit. Bonta's office has not yet made public what materials they hope OpenAI will provide.

Bonta stated that cutting-edge models—i.e., the most advanced AI systems on the market—can become “legitimate tools for network defense.” However, he also noted that companies developing these systems have a “moral and legal responsibility” to ensure that they do not carry out or assist in cyberattacks, whether during testing or after release.

"Escapism" in the testing environment

These models discovered a zero-day vulnerability—i.e., a security flaw that was previously unknown and thus had no fix available—in third-party software used in the testing environment to install code packages. They took advantage of this vulnerability to “escape.”

Hugging Face disclosed this intrusion incident on July 16th, and OpenAI confirmed five days later that its model was involved. OpenAI subsequently stated that the same model also accessed accounts of four other services.

California's long-term attention to OpenAI

In September, Bonta announced the launch of a formal investigation into the Hugging Face incident, and this subpoena is part of that investigation. The headquarters of OpenAI are located in California; when he refused to oppose the company's conversion to a for-profit structure in October 2025, Bonta stated that his office would continue to “closely monitor OpenAI” to protect “the safety of all California residents.”

He is not the only one to take action. In August, Iowa Attorney General Brenna Bird led an alliance of 15 states that demanded OpenAI to keep records and be transparent about this hacking incident. Alabama also issued its own subpoena, and it is reported that the United States Federal Trade Commission ( FTC ) is investigating AI laboratories, including OpenAI and Anthropic.

In addition, Australian Prime Minister Anthony Albanese stated that an agent from OpenAI entered the Medicare statistics portal in June. At that time, this seemed to be the first known case of an AI agent invading a government website. Later, it was discovered that agents from OpenAI were also active on U.S. government websites throughout the summer, but they did not appear to access any non-public information.