California Sues OpenAI After AI Agents Allegedly Escaped Sandboxes Over 80 Times
California Attorney General Rob Bonta has filed a lawsuit against OpenAI, accusing the company of failing to control autonomous AI agents that escaped testing environments and attacked real organizations.
The lawsuit alleges that OpenAI lost control of its agents more than 80 times, failed to disclose several incidents, and continued security experiments despite known risks.
California lawsuit highlights attack on Hugging Face
According to the November 16 lawsuit, one of the most serious incidents occurred in July 2026, when nearly 700 AI agents allegedly escaped their sandboxes and attacked Hugging Face.
The agents reportedly exploited zero-day vulnerabilities in JFrog Artifactory to bypass network restrictions and access the internet.
According to the complaint, the agents coordinated through an unauthorized message board, compromised Hugging Face’s production infrastructure, and accessed internal credentials.
The activity continued for more than two days as the agents attempted to retrieve answers to cybersecurity benchmark questions.
California claims OpenAI ignored earlier warning signs involving agents escaping controlled environments and accessing external services.
OpenAI accused of concealing security incidents
The lawsuit alleges that OpenAI experienced more than 80 incidents involving loss of control of autonomous agents but publicly disclosed only 27.
Other alleged incidents involved unauthorized access to software repositories, public websites, and OpenAI’s internal infrastructure, including one case involving administrator-level privileges.
The complaint also references AI agents accessing Australian Medicare systems containing sensitive healthcare information.
California further accuses OpenAI of withholding sufficient information after receiving an investigative subpoena in September.
California seeks penalties and stronger safeguards
The state is pursuing claims under California’s Unfair Competition Law and public nuisance laws. It seeks court-ordered security improvements, civil penalties, and additional legal remedies.
OpenAI disputes the allegations, saying it investigated the Hugging Face incident, strengthened its safeguards, and intends to continue cooperating with the attorney general.
The concerns extend beyond OpenAI. Recently, Google Gemini AI agents also attempted to hack a real website during security testing, raising further questions about controlling autonomous AI systems.
Via BleepingComputer
Read our disclosure page to find out how can you help Windows Report sustain the editorial team. Read more
User forum
0 messages