California attorney general subpoenas OpenAI amid probe of AI model security risks

California Attorney General Rob Bonta has issued an investigative subpoena to OpenAI, demanding documents and testimony as part of a broader probe into the security of the company’s artificial‑intelligence models.

Scope of the Subpoena

The subpoena seeks information on OpenAI’s internal security protocols, incident‑response procedures, and any known breaches or vulnerabilities affecting its flagship models, including ChatGPT and the newer Gemini‑style system.

Why California Is Investigating

State regulators have grown increasingly concerned that AI systems could be exploited to launch phishing attacks, generate deepfakes, or facilitate ransomware campaigns, potentially exposing California residents and businesses to heightened cyber risk.

OpenAI’s Response

OpenAI confirmed receipt of the subpoena and said it will cooperate fully while also reviewing the request for relevance and confidentiality. The company has previously highlighted its own “Red Team” testing program aimed at identifying model weaknesses.

• Review of model training data for inadvertent inclusion of malicious code
• Assessment of API access controls and rate‑limiting mechanisms
• Documentation of past security incidents involving AI‑generated content

Legal experts note that the subpoena could set a precedent for state‑level oversight of AI safety, adding pressure on the industry to adopt more transparent security standards.

The rapid deployment of powerful language models demands rigorous oversight to protect public safety, said a cybersecurity analyst familiar with the case.

If the investigation uncovers significant gaps, California may pursue further enforcement actions, including fines or mandatory remediation plans, echoing earlier actions taken against tech firms over data‑privacy violations.

For more details, see Reuters coverage of California AG subpoena to OpenAI.