California Attorney General serves a subpoena on OpenAI over cybersecurity
IA en un minuto newsroom · Editor: Jon Elgezabal

In 30 seconds
Rob Bonta, the top law officer in California, has sent OpenAI a formal demand to answer further questions about its cybersecurity. It was delivered on September 30, announced the next day and added to the probe opened in September over what happened at Hugging Face. Bonta stresses that his office is looking into the facts and is not accusing anyone.
California Attorney General Rob Bonta has served OpenAI with an investigative subpoena, a formal demand that the company respond to additional questions about its cybersecurity incidents and risks.
His office served it on September 30 and announced it from Oakland on October 1. It forms part of the formal investigation into the Hugging Face incident that the office announced in September (Hugging Face is the platform where open AI models are shared), and of a wider review of cybersecurity incidents and risks affecting the company and its models. The press release does not detail what happened in that incident, and the office is asking anyone with information about this case or similar ones to come forward.
Bonta said the most advanced models "can be legitimate tools for cyber defense," but that the companies developing and offering them are morally and legally responsible for making sure they neither carry out nor enable cyberattacks, whether in testing or once in service. Developers that fall short "can and should be held legally accountable," he said, adding that his office intends to establish whether that is the case here. It is an investigation, not an accusation.
In September, a bipartisan group of attorneys general that included Bonta had written to Congress urging it to regulate large AI models and their developers without delay, following reports of serious cybersecurity incidents at several leading labs. And in January he announced a separate investigation into reports of nonconsensual sexually explicit material generated with Grok, the xAI model, and circulated on X.
Why it matters · analysis and opinion
Is a company responsible for what its model does while it is still testing it? California's Attorney General argues that it is, even before the model reaches the public, and that idea goes beyond this particular case. For now the subpoena only requires OpenAI to answer questions: there is no accusation, and the office has not said exactly what happened at Hugging Face. It remains to be seen what it finds and whether that view ends up turned into concrete requirements. In the meantime, the message reaches any lab training models capable of attacking systems, and indirectly the companies that hire them. Asking a provider how it monitors its tests and what it does when something slips through is no longer an exotic question.
Official source: California Attorney General · Written with the help of AI: how we make the news


