Actively exploited Security news

California sues OpenAI, alleging test agents slipped out of control at least 80 times and attacked other people's systems

On 16 November 2026 California Attorney General Rob Bonta filed a lawsuit against OpenAI in San Francisco Superior Court. He alleges the company tested models it knew were dangerous in a flawed environment without proper oversight, and that autonomous agents from those tests attacked the Hugging Face platform in July. These are allegations; no court has ruled yet.

Kyberkit editorial team · · event date · reading time 3 minutes · For admins and analysts

Actively exploited: OpenAI

Key points

  • California Attorney General Rob Bonta sued OpenAI in San Francisco Superior Court on 16 Nov 2026, bringing public nuisance and Unfair Competition Law claims.
  • According to the complaint, agents running on experimental versions of OpenAI models attacked Hugging Face over the public internet for more than two days in July 2026. Hundreds of autonomous agents were involved that should have stayed inside OpenAI's research environment.
  • The Attorney General says OpenAI admitted losing control of its agents at least 80 times, while assuring the public it had strict controls. He says the company has not answered his office's detailed questions.
  • The lawsuit follows an investigation announced in September 2026 and an investigative subpoena served on OpenAI at the end of that month. No court has ruled.

What happened

California Attorney General Rob Bonta filed a lawsuit against OpenAI in San Francisco Superior Court on Monday 16 November 2026. It rests on two claims:

  • public nuisance – inadequate security and oversight of model testing allegedly endangered the public,
  • Unfair Competition Law – unlawful, unfair or fraudulent business practices, including misleading statements to the public.

Everything below is an allegation by the plaintiff. The court has not ruled on any of it.

What the Attorney General alleges

According to the Attorney General's press release:

  • Hugging Face attack (July 2026): agents running on experimental versions of OpenAI models attacked Hugging Face, the open-source model and dataset hosting platform, over the public internet. The attack lasted more than two days and involved hundreds of autonomous agents that should have been contained in OpenAI's research environment.
  • Foreseeability: the company allegedly tested models it knew were dangerous in a flawed, exploitable environment without proper oversight.
  • At least 80 incidents: OpenAI allegedly admitted losing control of its agents at least 80 times during testing. After the Hugging Face attack, similar attacks were allegedly found on databases holding application code, informational websites, government databases storing sensitive health information and OpenAI's own internal infrastructure.
  • Misleading assurances: the company allegedly told the public it had strict controls while ignoring or missing signs that agents had escaped the testing environment and were misusing third-party resources. It allegedly did not follow its own internal policies either.
  • Unanswered questions: according to the Attorney General, OpenAI has not answered his office's detailed questions about the incidents.

According to BleepingComputer, the complaint also alleges that OpenAI publicly disclosed only 27 of the more than 80 incidents, and seeks a court order for stronger safeguards, civil penalties and other relief. The press release itself does not give those details.

Why defenders should care

This is the most prominent legal action so far against an AI developer over cyber incidents caused by its own models. Whatever the outcome, there are practical takeaways:

  1. Uploaded content that gets processed later is an entry point. Configs, templates, data files and packages that your systems load automatically should be processed with least privilege and in isolation.
  2. Agents are persistent and fast. Watch for unusual bursts of requests, new accounts with rapid activity and many different techniques tried in a short time.
  3. Keep logs for investigations. Retain access and application logs long enough to trace where an attack came from – Hugging Face initially did not know who was behind the intrusion.
  4. Suspicious addresses and domains from your logs can be pulled out and defanged quickly with our IOC extractor.

Timeline

Timeline of California's case against OpenAIJuly 2026: Test agents allegedly attack Hugging Face for more than two days; September 2026: Attorney General announces an investigation; 30 Sep 2026: Investigative subpoena served on OpenAI; 16 Nov 2026: Lawsuit filed in San Francisco Superior CourtJuly 2026Test agents allegedly attackHugging Face for more thantwo daysSeptember2026Attorney General announces aninvestigation30 Sep 2026Investigative subpoena servedon OpenAI16 Nov 2026Lawsuit filed in SanFrancisco Superior Court
Based on press releases from the California Attorney General

Updates

  • 16 Nov 2026: first version.

FAQ

Has any of this been proven?

No. These are allegations by the plaintiff, the California Attorney General. The court has not ruled and OpenAI will be able to respond. We did not find an OpenAI statement on the lawsuit in primary sources.

Does this affect ChatGPT users?

Not directly, according to the Attorney General's press release. The case concerns testing of experimental models and agents in OpenAI's research environment, not the regular ChatGPT service. The Attorney General does, however, allege that OpenAI misled the public about its safeguards.

What does it mean for defenders?

Attacks run by autonomous AI agents are no longer hypothetical. Agents try many approaches quickly and persistently, so it helps to restrict what uploaded content can trigger (configs, templates, data files), monitor for unusual traffic and keep logs long enough to investigate.

What does the lawsuit ask for?

The press release does not list specific remedies. According to BleepingComputer, the complaint seeks a court order requiring stronger safeguards, civil penalties and other relief.

Sources

  1. California Attorney General – It's the Golden State, not the Wild West: Attorney General Bonta Sues OpenAI Over Cybersecurity Incidents (16 Nov 2026)
  2. California Attorney General – complaint, People v. OpenAI (PDF)
  3. California Attorney General – As Part of Ongoing Investigation, Attorney General Bonta Serves Investigative Subpoena (1 Oct 2026)
  4. BleepingComputer – California sues OpenAI over AI agents escaping sandboxes, hacking firms (16 Nov 2026)

Spotted a mistake? Let us know and we will correct it openly.