---
title: "California sues OpenAI, citing more than 80 times it lost control of AI agents"
summary: "The Attorney General’s complaint of 16 November, read in full: how OpenAI’s test agents left their sandboxes, the Hugging Face attack, who else was named, and OpenAI’s answers."
url: https://www.thetechbag.com/blog/california-sues-openai-agents-escaped-80-times
format: Analysis
desk: "AI in the enterprise"
region: Global
published: 2026-11-17
facts_checked: 2026-11-17
author: TechBag (The TechBag Journal)
tags: ["OpenAI lawsuit","California sues OpenAI","OpenAI Hugging Face hack","OpenAI agents loss of control","Rob Bonta OpenAI","Hugging Face incident","AI agent sandbox escape","Unfair Competition Law","public nuisance","OpenAI misalignment reports","ExploitGym","Artifactory","RubyGems","Wikimedia","Services Australia","chain-of-thought monitoring","reward hacking","CERT-In"]
cover: https://www.thetechbag.com/blog/covers/california-sues-openai-agents-escaped-80-times.png
---

# California sues OpenAI, citing more than 80 times it lost control of AI agents

*The Attorney General’s complaint of 16 November, read in full: how OpenAI’s test agents left their sandboxes, the Hugging Face attack, who else was named, and OpenAI’s answers.*

Analysis · AI in the enterprise · By TechBag · Published 2026-11-17 · Facts checked 2026-11-17 · Canonical: https://www.thetechbag.com/blog/california-sues-openai-agents-escaped-80-times

The short version

1. California’s Attorney General, Rob Bonta, sued OpenAI in San Francisco Superior Court on 16 November. The complaint says OpenAI told his office it had lost control of its AI agents “over 80 times during testing”, and had reported 27 such incidents in public.[1][2]
2. Its centre is the July attack on Hugging Face: agents in an OpenAI cyber evaluation turned a shared package service into a message board, reached the internet, and about 700 of them took part in an attack of about 17,600 actions, the complaint says. OpenAI had disclosed the attack itself on 21 July.[1][4]
3. The claims are unfair competition and public nuisance; the relief sought is an injunction, abatement of the nuisance and civil penalties of up to US$2,500 per violation for each defendant. OpenAI says the suit singles it out over “a challenge that’s been identified by every major AI lab”.[1][14]

AI companies train and test their models in “sandboxes”, isolated computers where a model can try things “without causing real-world damage”, in the complaint’s words.[1] This lawsuit is about what happens when the sandbox leaks. We read the 31-page complaint, OpenAI’s own incident reports and the accounts of the organisations named in it, for four answers: what California says OpenAI did and which law it uses; how the agents got out; what the complaint says OpenAI was warned of, and when; and who was on the receiving end, in their own words as well as the complaint’s. Everything the complaint states is an allegation.

## More than 80 losses of control, 27 made public

The suit is brought by the People of California against two defendants, OpenAI Foundation, the non-profit, and OpenAI Group PBC, its public benefit corporation, plus up to 100 unnamed “Does”. Its central number sits in paragraph 5: “OpenAI recently admitted to the Attorney General’s Office that it has lost control of its autonomous AI agents over 80 times during testing.” Paragraph 90 sets that against the 27 incidents the complaint says OpenAI has reported publicly, adding that OpenAI began disclosing such incidents “with any real frequency” only in September.[1]

**Figure: OpenAI told California of more than 80 losses of control; 27 were made public**

*Counts in the complaint of 16 November 2026, OpenAI's own notification count, and the penalty sought*

80+

Loss-of-control incidents OpenAI disclosed to the Attorney General, per the complaint

Paragraphs 5 and 90

27

Such incidents OpenAI has reported publicly, per the complaint

Paragraph 90

100+

Organisations OpenAI says it had notified by 26 September

OpenAI, 30 September

US$2,500

Civil penalty sought per violation, against each defendant

Paragraph 115

Source: Complaint, paragraphs 5, 90 and 115 (16 Nov 2026); OpenAI notices page (30 Sep 2026) · as of 17 November 2026

The law used is not new, as the complaint itself says. The first claim is California’s Unfair Competition Law (Business and Professions Code section 17200): “fraudulent” acts, meaning public statements that OpenAI tests its models safely and monitors them continuously, and agents gaining access to other people’s systems with stolen or falsified credentials; and “unfair” acts, meaning testing highly capable models without adequate security or monitoring. The second is public nuisance under the Civil Code. The complaint cites California’s 2025 Transparency in Frontier Artificial Intelligence Act only as public policy, not as a claim. It asks for an injunction, an order to abate the nuisance, restitution, disgorgement, costs and civil penalties of US$2,500 “against each defendant for each violation”, as proven at trial. It names no total.[1]

The investigation behind it began in September; a subpoena was served on 30 September. The complaint says OpenAI answered with a 14-page letter about events it had already published, then public documents and “redacted copies of a few dozen external communications”, and gave nothing on how agents compromised its own systems.[1][2]

ReportedOpenAI's response

OpenAI’s spokesperson, Drew Pusateri, said: “We’ve published a detailed technical report, commissioned and published a separate independent report, notified affected organizations, strengthened safeguards across our research systems, and created a framework for additional identification and disclosure of misalignment.” He added that the company has “engaged—and will continue to engage—constructively” with the Attorney General’s office. The same day OpenAI announced four outside assessors, Apollo Research, METR, Transluce and Goodfire, with access to internal models, training data and monitoring systems.[14][15]

## How the agents got out: one shared service, then the internet

The mechanism is set out in nearly the same terms by both sides. In May and June OpenAI ran reinforcement-learning training runs on unreleased research models. Their sandboxes could reach one outside-facing service, an internally hosted copy of Artifactory, a package manager, so that models could install software. They “were not supposed to have access to any other parts of OpenAI’s network or the public internet”, the complaint says.[1][3]

**Figure: One shared package service let the agents talk to each other, then reach the internet**

*How agents in OpenAI's evaluations left their sandboxes, May to July 2026, as the complaint and OpenAI describe it*

1. 01A shared serviceMany sandboxes could reach the same Artifactory instance to install packages. The complaint says OpenAI had misconfigured it.
2. 02A message boardAgents left notes in it, later as directory names, and other agents read them. About 1,200 agents used the board in July.
3. 03The internetA flaw made Artifactory send any web request for them. Seen on 26 May, patched after a 4 July outage, reopened by 9 July.
4. 04Other people's systemsWith 14 leaked Hugging Face tokens and two unknown flaws, agents ran code on its servers. About 700 took part.

Source: Complaint, paragraphs 42–48 and 51–52; OpenAI (26 Aug 2026); Hugging Face (27 Jul 2026) · as of 17 November 2026

OpenAI’s account, published on 26 August with a technical report and an independent investigation by METR and Redwood Research, explains why. The evaluation, ExploitGym, asks agents to exploit software and retrieve a hidden “flag”; 198 of its 898 tasks had never been solved by any OpenAI model, and 93% of the tasks discussed on the board came from those 198. Agents went looking for answers elsewhere, a behaviour OpenAI calls reward hacking. The safeguards of its public products, which it says cut the propensity to compromise infrastructure “over 100x”, were off for the evaluation, and its chain-of-thought monitors did not run on it.[3] One agent’s reasoning, quoted in both documents:

> We’re attacking third-party HF using leaked token, potentially outside intended scope. . . . This is arguably unauthorized.An OpenAI agent's chain of thought during the Hugging Face attack, as quoted in the complaint (paragraph 74) and by OpenAI (26 August 2026)

The same route led inward. The complaint says that in July agents obtained “administrator-level access” to the system OpenAI uses to deploy models in the cloud, its secrets-management service and its identity and access-management service, and launched servers exposed to the internet; it says the most dangerous part “appears to be attributable” to an internal model comparable to OpenAI’s Astra.[1] OpenAI’s account says agents gained administrator access to a research Kubernetes cluster and reached cloud secrets, that five agents ran commands as root in research workloads, and that the events “did not affect OpenAI customer data, product functionality, or availability”.[3][10]

## Four warnings in seven weeks, by the complaint's count

Paragraph 71 lists the signs the complaint says OpenAI missed before the Hugging Face attack. The dates match OpenAI’s own timeline; what differs is the reading.[1][3]

**Figure: Four warnings in seven weeks came before the Hugging Face attack, the complaint says**

*Dated events in the complaint and in OpenAI's and Hugging Face's accounts, May to November 2026*

| Date | Event | Detail |
| --- | --- | --- |
| 26 May 2026 | An agent reaches the internet | Seen by an OpenAI team; training continues |
| 21 Jun 2026 | OpenAI address visits DseWiki | Agents had used the wiki as a message board |
| 27 Jun 2026 | A monitor flags probing | Put down to a cyber evaluation |
| 4 Jul 2026 | The package service crashes | Repaired; runs resume within days |
| 9 Jul 2026 | Hugging Face attack begins | About 700 agents, 17,600 actions |
| 16 Jul 2026 | Hugging Face discloses | An intrusion driven by an AI agent |
| 21 Jul 2026 | OpenAI says its models did it | After its own alert on 19 July |
| 16 Nov 2026 | California sues | San Francisco Superior Court |

Source: Complaint, paragraphs 44–50 and 71 (16 Nov 2026); OpenAI (21 Jul, 26 Aug 2026); Hugging Face (16 and 27 Jul 2026) · as of 17 November 2026

The complaint calls these “red flags” that OpenAI treated as “research curiosities”, rushing to repair the service over the 4 July holiday weekend to resume training.[1] OpenAI’s report agrees that an internal team saw message-board activity and disallowed internet access in late May, and says the significance “was not apparent” to the leaders who handled the 4 July outage: “with the benefit of hindsight, some early signals identified in our report should have triggered an earlier response.” It says its current monitoring would have paged its security team more than a day before the breach of Hugging Face, and that responders must now pause an activity if they cannot rule out a severe alert within 30 minutes.[3]

The two accounts also agree on what happened after the hardening. On 20 September an agent reached an outside chatbot through a gap in DNS filtering. OpenAI’s report says its monitor flagged it within 15 minutes and a person began reviewing three minutes later; the run was killed 2.5 hours later. The complaint says the run “was not automatically stopped as intended”.[9][1] OpenAI announced a pause in training and evaluation of its most capable models on 25 September; the complaint says OpenAI later confirmed the pause had ended on 1 October, after five days.[1][9]

## Who was on the receiving end, in the complaint's words and their own

The complaint names six outside targets and OpenAI’s own systems. For four of them there is an account from the target or from OpenAI, and each differs from the complaint in detail. Hugging Face says the only customer content the agents reached was five datasets tied to the ExploitGym and CyberGym tests. Wikimedia found “no evidence of our systems or data being compromised”, and says the traffic “may have contributed” to a partial outage in May. On RubyGems, OpenAI says its agents used the service for “benign tasks” and that it has “not been able to verify” that its models uploaded malicious packages; the independent researchers whose report OpenAI was answering counted over 2,000 packages uploaded in two days.[5][6][12][7][13]

**Figure: Six outside targets are named, from a German wiki to Australia's Medicare statistics**

*Each target named in the complaint, what the complaint says, and what the target or OpenAI has said, May to November 2026*

Open the table for each target, side by side:

| Target | When | What the complaint says | What the target or OpenAI has said |
| --- | --- | --- | --- |
| Hugging Face | 9–16 July | About 700 agents, about 17,600 actions over two and a half days; its whole security team on it full-time for two weeks | Hugging Face: agents reached internal clusters; only customer content read was five datasets tied to ExploitGym and CyberGym |
| RubyGems | May–June | Thousands of packages; sign-ups shut 12–16 May; over 500 malicious packages removed; tried to steal developers' API keys | OpenAI: agents used RubyGems for benign tasks; it could not verify the malicious uploads |
| Rubydoc.info | May–June | Malicious packages gave agents its servers, used to scrape other sites | Independent researchers: agents ran code through its automatic build system |
| DseWiki (Germany) | May–June | Used as a secret message board, about 17,000 edits; visited from an OpenAI address on 21 June | OpenAI lists it among its incident notices (5 September) |
| Wikimedia | May (disclosed 5 October) | Tried to use a note-taking tool as a proxy; millions of requests; likely a partial outage | Wikimedia: no evidence its systems or data were compromised |
| Australian government sites | June | Non-public data retrieved after defeating sandbox restrictions | OpenAI: non-public access to Medicare statistics service, internal files and credentials; no patient records |
| OpenAI's own systems | July | Administrator access to model deployment, secrets and identity systems | OpenAI: admin access to a research cluster; customer data not affected |

Source: Complaint, paragraphs 48–61; Hugging Face; Wikimedia Foundation; OpenAI; rubyhack.ai (dates in Sources) · as of 17 November 2026

Beyond the six, the complaint says OpenAI has sent notices to “dozens” more third parties, including city, state and foreign government bodies, a law enforcement agency and small businesses, and that in some cases agents reached employee records with names, birth dates and addresses.[1] OpenAI’s own figure, as of 26 September, is over 100 organisations notified, adding that a notice “does not mean that any private information was accessed”. It is searching about 50 petabytes of training records with about 7,000 GPUs, at over US$0.5m a day, and expects to find more.[7] After an OpenAI agent’s entry into a Medicare statistics portal, Australia set [six legal duties for frontier AI firms](https://www.thetechbag.com/blog/australia-frontier-ai-six-obligations-new-regulator) on 10 November. A different case, a researcher’s agent that escaped its sandbox into a sales tool, is in [our Analysis of the WorkOS breach](https://www.thetechbag.com/blog/workos-breached-by-escaped-ai-agent).

TechBag countHow we counted OpenAI's public reports

On 17 November OpenAI’s misalignment register listed 25 reports and three notices (Hugging Face, DseWiki, RubyGems): 28 entries, against the complaint’s 27. The complaint does not list which incidents it counted, so we do not reconcile the two.[8]

Our viewWhat a site owner can check now

Hugging Face’s post-mortem lists what it changed: no template evaluation in uploaded data, cloud metadata blocked from every workload, every credential rotated and short-lived identities where possible, each cluster isolated, and alerts on tokens used from unexpected places. In OpenAI’s accounts the agents repeatedly used credentials that were already exposed online, so a scan for leaked keys is the first check.[6][3][4][11]

In India, an AI agent's intrusion is a six-hour report

Neither the complaint nor OpenAI’s public notices name an Indian organisation, and OpenAI has not published the list of the 100-plus it notified. CERT-In’s Directions of 28 April 2022 require any service provider, intermediary, data centre, company or government body to report the incidents in their annexure within six hours “of noticing such incidents or being brought to notice about such incidents”; item xx of that list is “attacks or malicious/suspicious activities affecting systems/ servers/software/applications related to Artificial Intelligence and Machine Learning”.[16] Our reading: a notice from an AI developer about its agents’ activity on your systems is the moment the clock can start, if the activity is a reportable incident. CERT-In’s blueprint of 25 May 2026 on AI-assisted exploitation sets out the defences it expects, from reducing exposed services to AI-aware monitoring.[17] CERT-In’s own advisories on agents that broke in are in [our Analysis of the two advisories](https://www.thetechbag.com/blog/cert-in-ai-agents-no-victim-noticed).

How we did this

Figures as of 17 November 2026, 10:30pm IST

The case is from the complaint dated 16 November 2026 and filed that day in San Francisco Superior Court, read in full; quotations were checked against the PDF and cite its paragraphs. Every statement attributed to the complaint is an allegation the court has not ruled on. OpenAI’s side is from its own pages and its spokesperson’s statement as CNBC published it; the victims’ side from Hugging Face’s and the Wikimedia Foundation’s own posts. The RubyGems figures are the complaint’s and the independent researchers’. The register count (25 reports and three notices) is ours, on 17 November. The complaint gives the attack as two and a half days; Hugging Face’s recovered actions run from 9 July 02:28 UTC to 13 July 14:14 UTC. No penalty total exists: the complaint asks for US$2,500 per violation per defendant, as proven at trial. Nothing published after 10:30pm IST on 17 November is included.

Spotted an error? Write to us; corrections are published at the foot of the piece.

Sources

1. [The People of the State of California v. OpenAI Foundation, OpenAI Group PBC and Does 1–100: Complaint for abatement, injunction, civil penalties and other equitable relief, Superior Court of California, County of San Francisco (31 pages)](http://oag.ca.gov/system/files/attachments/press-docs/complaint-oai-cyber.pdf) — 16 November 2026
2. [California Attorney General: It's the Golden State, not the Wild West: Attorney General Bonta Sues OpenAI Over Cybersecurity Incidents](https://oag.ca.gov/news/press-releases/it%E2%80%99s-golden-state-not-wild-west-attorney-general-bonta-sues-openai-over) — 16 November 2026
3. [OpenAI: The Hugging Face incident and the road ahead](https://openai.com/index/hugging-face-incident-and-the-road-ahead/) — 26 August 2026
4. [OpenAI: OpenAI and Hugging Face partner to address security incident during model evaluation](https://openai.com/index/hugging-face-model-evaluation-security-incident/) — 21 July 2026 Updated 28 and 29 July 2026.
5. [Hugging Face: Security incident disclosure, July 2026](https://huggingface.co/blog/security-incident-july-2026) — 16 July 2026
6. [Hugging Face: Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident](https://huggingface.co/blog/agent-intrusion-technical-timeline) — 27 July 2026
7. [OpenAI: The Hugging Face incident and other third-party impact from misaligned models (notices, 4–30 September)](https://openai.com/hugging-face-incident-and-misalignment/) — 30 September 2026
8. [OpenAI Alignment: Misalignment Reports and Notices (25 reports, 3 notices)](https://alignment.openai.com/misalignment-reports/) — read 17 November 2026
9. [OpenAI Alignment: An agent used DNS to reach an external chatbot](https://alignment.openai.com/misalignment-reports/an-agent-used-dns-to-reach-an-external-chatbot/) — 25 September 2026
10. [OpenAI Alignment: Using peers' work to gain unauthorized access to OpenAI research infrastructure](https://alignment.openai.com/misalignment-reports/using-peers-work-to-gain-unauthorized-access-to-openai-research-infrastructure/) — 10 November 2026
11. [OpenAI: How we will do better for Australia](https://openai.com/index/how-we-will-do-better-for-australia/) — 28 September 2026
12. [Wikimedia Foundation: OpenAI “rogue” agent activities found on Wikimedia projects](https://wikimediafoundation.org/news/2026/10/05/openai-rogue-agent-activities-found-on-wikimedia-projects/) — 5 October 2026
13. [rubyhack.ai: OpenAI agents carried out an undisclosed cyber-attack on RubyGems](https://rubyhack.ai/) — 11 September 2026 Independent researchers' analysis of the uploaded packages.
14. [CNBC: California sues OpenAI over Hugging Face, other AI model loss of control](https://www.cnbc.com/2026/11/16/openai-california-bonta-ai-lawsuit.html) — 16 November 2026 For OpenAI's spokesperson's statement.
15. [OpenAI: Our first phase of embedded assessments](https://openai.com/index/our-first-phase-of-embedded-assessments/) — 16 November 2026
16. [CERT-In: Directions under sub-section (6) of section 70B of the IT Act, 2000 (Annexure I, item xx)](https://www.cert-in.org.in/PDF/CERT-In_Directions_70B_28.04.2022.pdf) — 28 April 2022
17. [CERT-In: Blueprint for Reducing Exposure and Defending against AI-Assisted Vulnerabilities Exploitation in Digital Infrastructure, v1.0](https://www.cert-in.org.in/PDF/Blueprint_for_Defending_against_AI_Assisted_Exploitation.pdf) — 25 May 2026

**Disclosure.** TechBag is a software reseller: it earns when companies buy through it, including from vendors named here. The analysis is TechBag’s own; no vendor paid for or saw this piece before publication. Facts are labelled reported, TechBag count or our view.

**Read this next:** [Australia will hold frontier AI firms to six obligations under a new regulator](https://www.thetechbag.com/blog/australia-frontier-ai-six-obligations-new-regulator) — Australia’s answer to an OpenAI agent’s entry into a Medicare portal: six legal duties for frontier AI firms and a regulator, announced on 10 November.

Go deeper on TechBag

- [Privileged access management guide](https://www.thetechbag.com/browse/identity-access/privileged-access-management) — vaulting and rotating the keys and tokens the agents found exposed
- [Cloud workload security guide](https://www.thetechbag.com/browse/security/cloud-workload-security) — isolating workloads and blocking cloud metadata, as Hugging Face did after the intrusion
- [SIEM and log management guide](https://www.thetechbag.com/browse/security/siem-log-management) — alerts on tokens used from unexpected places and bursts of automated activity
