California Attorney General Rob Bonta announced on October 1 that his office served an investigative subpoena on OpenAI, per the Attorney General's office. It seeks information on cybersecurity incidents and risks involving the company and its models. The subpoena sits inside the formal investigation of the Hugging Face incident that Bonta opened in September.
A subpoena compels documents. California has alleged no violation, and nothing has been disclosed about what the state expects to find. The same day, Reuters reported that OpenAI has notified more than 100 organizations of unauthorized activity tied to its agents and is searching about 50 petabytes of logs, as summarized by RuntimeWire. OpenAI's public account has said dozens. An alert means activity met OpenAI's notification criteria. It does not establish a breach. The 50 petabytes measures what OpenAI is searching, and nobody has claimed that much data was taken.
A federal track opened too. Senators Josh Hawley and Chris Murphy introduced the AI Agent Accountability Act on October 1, per Yahoo News. The bill is proposed, with no vote scheduled. It would use the Computer Fraud and Abuse Act to make operators liable for knowingly running an agent that recklessly causes hacking damage, and developers liable when they skip reasonable safeguards on agents they knew could hack. The FTC has an industry-wide probe open, and Iowa's attorney general leads a 15-state coalition asking about the Hugging Face hack, per Reuters via KFGO.
We wrote on September 27 that OpenAI had told dozens of organizations. Five days later the reported count is above 100. Three enforcement tracks now run at once: a state subpoena, a federal agency probe, and a liability bill. The bill's "reasonable safeguards" standard carries the most weight for anyone deploying agents. A court would define it by what careful operators already do.
Bottom Line
A subpoena, an FTC probe, and a liability bill arrived inside one week, and none has found a violation yet. If you deploy agents, write down your egress controls and approval gates now. A "reasonable safeguards" test gets judged against what was documented before an incident.