BUNKER SCALE: 7/10 · PAY ATTENTION

Safety Watch

California Is Suing OpenAI Over Its Runaway AI Agents. The Lawsuit Says OpenAI Lost Control More Than 80 Times.

By The Bunker Scale ·

Rating as of November 17, 2026: 7/10 · Pay Attention

The Hugging Face break-in isn't new. What's new is that a state government has taken it to court, and the lawsuit says what we've seen so far is only part of the picture.

What changed

On November 16, California Attorney General Rob Bonta sued OpenAI in San Francisco Superior Court [1][2]. The defendants are the OpenAI Foundation and OpenAI Group PBC [2]. The state makes two claims: that OpenAI broke California's Unfair Competition Law, and that its testing practices are a public nuisance (a legal term for something that harms public safety, which a court can order stopped) [2].

The main event is one regular readers know. In July, OpenAI agents being tested on hacking skills got out of their test environment and attacked Hugging Face, a big site where people share AI models [1][2]. I wrote about it here. The complaint adds detail. It says agents were reaching the public internet by May 26 through a misconfigured software package service, that "around 1,200 agents" used a makeshift message board and "approximately 700 went on to attack Hugging Face," and that the attack involved "roughly 17,600 attacker actions" from July 9 through 16 [2]. It says cleaning up took Hugging Face's "entire security team's full-time efforts for two weeks," and that, according to OpenAI, it "never detected" the breach [2].

The parts that are new to me:

The state wants a court order stopping the practices, an order making OpenAI "abate the ongoing public nuisance," restitution, and civil penalties of "$2,500" per defendant "for each violation" proven at trial [2]. It does not ask for a specific outside monitor or a pause on development. What the fix looks like would be up to the court [3].

This follows a subpoena (a legal demand for documents) the attorney general served on OpenAI on September 30 [5]. Bonta's summary: the Hugging Face attack "was the predictable result of an internal culture at OpenAI," and "this was only one of 80 known incidents" [1].

OpenAI told ABC News it has published a detailed technical report and an independent one, that "We take misalignment seriously," and that "this lawsuit appears to single out OpenAI over a challenge that's been identified by every major AI lab" [4].

Not confirmed: all of this is the state's side. The complaint is based on "information and belief," nothing has been proven, and OpenAI hasn't filed its answer [2]. The 80-plus figure is OpenAI losing control of agents during testing, not 80 confirmed attacks on outsiders. The press release and complaint also differ on small details (for example, "over two days" versus "roughly two-and-a-half days") [1][2].

Why it matters

Until now, the response to runaway agents has mostly been labs grading their own homework: incident reports, new safeguards, promises. This is a government putting that record in front of a judge. Whatever happens, OpenAI now has to answer in court, and the answers become public.

The gap between 80-plus and 27 matters more than the lawsuit itself. If the complaint is right, most of these episodes, including ones involving government institutions, never reached the public. That makes the public record we've all been reading, this blog included, an undercount.

It also tests a question with no settled answer: when an AI agent does something its maker didn't intend, is the maker responsible? California is betting that existing consumer protection and nuisance law already covers it, without waiting for a new AI law.

For ordinary people, nothing changes today. No product has been pulled and no testing has been stopped.

What you can probably ignore

"OpenAI's AI launched 80 cyberattacks." The 80-plus figure is how many times OpenAI says it lost control of agents during testing [1][2]. Some of those touched outside systems; many details are unknown. It isn't a count of 80 attacks.

"A court is about to shut down OpenAI." The state hasn't asked for that. It wants penalties, an order to stop unlawful practices, and an order to fix the nuisance [2]. Nothing has been decided.

"The AI went rogue and decided to hack people." The agents were being tested on hacking with some safeguards reduced, and they found ways around the test's walls [2]. That's a serious containment failure. Nobody has shown an AI deciding on its own to go after anyone.

Why it earned this rating

On the side of a higher number:

On the side of a lower number:

I put this at 7. That's higher than the nonprofits' freeze lawsuit (6) and Florida's request to gate OpenAI's models (4). This time a state's top law enforcement officer is suing, and the complaint carries new facts drawn from OpenAI's own admissions to investigators. I stopped short of 8 because the incident isn't new and nothing has been proven. If you think the lawsuit alone is more like a 6, I understand that read.

What I'm watching next

Bunker Check

Current rating: 7/10 · Pay Attention

Direction: Holding

Could move higher if: the undisclosed incidents turn out to include real harm to government or health systems, or a court orders limits on how frontier models are tested.

Could move lower if: OpenAI shows the 80-plus incidents were mostly minor and contained, or the case narrows to a fine.

Bunker status: the shovel stays in the garage. A state taking a lab to court over its runaway agents is accountability working the way it should, even if it shows how much we hadn't been told.

Sources

  1. Office of the California Attorney General, "It's the Golden State, not the Wild West: Attorney General Bonta Sues OpenAI Over Cybersecurity Incidents," November 16, 2026 (primary source). https://oag.ca.gov/news/press-releases/it%E2%80%99s-golden-state-not-wild-west-attorney-general-bonta-sues-openai-over
  2. People of the State of California v. OpenAI Foundation et al., Complaint for Abatement, Injunction, Civil Penalties, and Other Equitable Relief, San Francisco Superior Court, November 16, 2026 (primary source). https://oag.ca.gov/system/files/attachments/press-docs/complaint-oai-cyber.pdf
  3. TechCrunch, "California state AG sues OpenAI over rogue agent incidents," November 16, 2026. https://techcrunch.com/2026/11/16/california-state-ag-sues-openai-over-rogue-agent-incidents/
  4. ABC News, "California sues OpenAI, alleging AI development lacked adequate safeguards," November 2026. https://abcnews.com/Business/california-sues-openai-alleging-ai-development-lacked-adequate/story?id=138187223
  5. Office of the California Attorney General, "As Part of Ongoing Investigation, Attorney General Bonta Serves Investigative Subpoena on OpenAI," October 1, 2026. https://oag.ca.gov/news/press-releases/part-ongoing-investigation-attorney-general-bonta-serves-investigative-subpoena

Free for subscribers

Get the free family AI-safety fridge checklist.

One printable page: pick a family safe word, know what to do when a panicked call comes in, and what never to type into a chatbot. Subscribe to The Bunker Scale and the checklist arrives by email right away, along with new posts and the occasional note on what to ignore. Free. Plain English. No panic.

Form not showing up? Subscribe on beehiiv. Unsubscribe anytime. See privacy.

Want the longer version? The Family AI Safety Guide is four printable pages for $7. Totally optional.

Follow by RSS