Most privacy class actions against tech companies turn on a single question: what did the company tell users, and where? Vredenburgh v. OpenAI OpCo, LLC, No. 3:26-cv-10527, filed September 16, 2026 in the Northern District of California, is built almost entirely on that question. The complaint alleges that OpenAI sent real ChatGPT conversations to contractors at an outside staffing firm for human review under a program code-named “Project Lily,” and that its Terms of Use, Privacy Policy and model-training pages never said so. OpenAI has not yet responded. What follows looks at the eight claims as pleaded and at the pressure points each is likely to face; none of the allegations has been proven.
The Theory: An Omission, Not a Lie
The complaint does not say OpenAI promised no human would read chats. It says OpenAI described its data practices at length and left the human reviewers out. The Privacy Policy allegedly lists eleven kinds of outside recipients of personal data, none of them a labeling or evaluation vendor. The model-training page allegedly walks through “What the process looks like” with no human reader. The one page that mentions human review limits it to flagged content checked by “our team” for policy violations.
Omission cases live or die on two things: whether the undisclosed fact was material, and whether the company had a duty to disclose it. The complaint’s best evidence on both is internal to OpenAI’s own documents. It points out that OpenAI plainly warns users when a workspace administrator can read a personal account’s chats, which the plaintiffs say shows OpenAI knew how to disclose human access and chose not to for training. It also points to Google, which the complaint says shows a human-review notice inside its chat interface.
OpenAI’s obvious answer is the Help Center FAQ that asks “Do humans view my content?” and says “trusted service providers” may access content “to improve model performance (unless you have opted out).” The case may come down to whether a disclosure that exists, but sits among about 45 nested help articles, counts as a disclosure a reasonable consumer would have seen.
Claim by Claim
1–3. Unfair Competition Law, Consumers Legal Remedies Act, False Advertising Law
California’s three consumer-protection statutes carry the omission theory. The pressure point is standing. The UCL and FAL require that a plaintiff lost money or property. Paying subscribers can say they paid a premium for a service whose privacy was misdescribed. Free users have to rely on the complaint’s second theory: that their prompts themselves had value, which the complaint supports by pointing to marketplaces where prompts are bought and sold. Courts have split on whether personal data counts as “property” lost for UCL purposes, and this is where free-user claims are most exposed.
The CLRA claim currently seeks only an injunction. That is the statute’s design: a plaintiff must give 30 days’ notice before seeking damages. The plaintiffs sent OpenAI a demand letter on September 16 and say they will add damages if it does not respond.
4. Fraudulent omission and concealment
The common-law version of the same theory, and the one that supports punitive damages. It requires intent to deceive and reliance, both of which get harder to show on a class-wide basis. Reliance on an omission is often presumed when the omitted fact is material, which is why the complaint works so hard on materiality.
5. Intrusion upon seclusion
The classic privacy tort requires an intrusion into a place or conversation where the plaintiff had a reasonable expectation of privacy, in a manner highly offensive to a reasonable person. The complaint’s most useful facts here are the ones about content: whole conversations, a history summary that can reveal a name or location, and users who asked ChatGPT to keep things confidential. The defense will argue users shared the content with OpenAI voluntarily and that a published FAQ defeats any expectation of privacy.
6. Invasion of privacy under the California Constitution
California’s constitutional privacy right requires a legally protected privacy interest, a reasonable expectation of privacy in the circumstances, and a serious invasion. It overlaps heavily with the intrusion claim and usually rises or falls with it.
7. California Consumer Privacy Act
This is the claim most likely to be challenged first. The CCPA’s private right of action is narrow: it covers unauthorized access, theft or disclosure of personal information resulting from a business’s failure to maintain reasonable security. The complaint frames it as OpenAI sharing personal information for an undisclosed purpose without reasonable safeguards. Whether a deliberate transfer to a vendor fits a statute written around security failures is an open question, and OpenAI will likely press it.
8. Unjust enrichment
California courts disagree about whether unjust enrichment is a standalone cause of action; many treat it as a quasi-contract claim for restitution. Its role here is to capture the value OpenAI allegedly gained from the reviewed prompts.
📨 Get Free Mass Tort Guides Alerts
Free · No spam · Unsubscribe anytime
The Remedies
The complaint seeks damages, restitution and punitive damages. The injunctive requests go further: separate opt-in consent before conversations go to outside reviewers, the “Improve the model for everyone” setting off by default, a warning in the chat window itself, and an order that OpenAI delete the reviewers’ work product and stop using, or retrain, any model built from it. That last request, a form of algorithmic disgorgement, is the most aggressive thing in the complaint.
The Class and the Early Motions
The proposed class is everyone in the United States who used ChatGPT, free or paid, with California, subscriber and California-subscriber subclasses. Enterprise, Business, Team, Edu and API users are excluded because they operate under separate agreements. A class this broad invites early challenges: whether individual questions about what each user read and expected overwhelm common ones, and whether each user’s settings history matters.
Before any of that, expect a motion to dismiss testing the claims above. Consumer tech cases often also see an early fight over whether the dispute belongs in individual arbitration under the company’s terms; OpenAI’s response, due October 13, 2026, will show whether it goes that way. The case is assigned to Magistrate Judge Alex G. Tse, who can preside only if all parties consent; otherwise it will be reassigned to a district judge. The first case management conference is December 18.
The privacy claims here echo the pleading patterns we traced in the LifeMD privacy settlement, while OpenAI’s separate exposure over training data is covered in the authors’ copyright class action. The complaint itself, with a running status summary, is at OpenClassActions.com, and the class and deadline tracker is at OpenClassActions.org.
Frequently Asked Questions
What does Vredenburgh v. OpenAI allege?
That OpenAI routed real ChatGPT prompts and conversations to outside contractors who read, summarized and scored them to improve its models, under a program code-named Project Lily, without disclosing it in the Terms of Use, Privacy Policy or model-training pages. OpenAI has not yet responded and the allegations are unproven.
What claims does the complaint bring?
Eight: California’s Unfair Competition Law, Consumers Legal Remedies Act and False Advertising Law; fraudulent omission and concealment; intrusion upon seclusion; invasion of privacy under the California Constitution; the California Consumer Privacy Act; and unjust enrichment.
Why does the CLRA claim seek only an injunction?
The Consumers Legal Remedies Act requires 30 days’ notice before a plaintiff seeks damages. The plaintiffs sent OpenAI a demand letter on September 16, 2026 and say they will add damages if OpenAI does not respond within 30 days.
Which claim is most vulnerable?
The CCPA claim faces the clearest challenge, because the statute’s private right of action is written around security failures rather than deliberate sharing. Free users’ UCL and FAL claims also face a standing question, since those statutes require lost money or property.
When is OpenAI’s response due?
October 13, 2026. The first case management conference is set for December 18, 2026, before Magistrate Judge Alex G. Tse.
Sources
- Class Action Complaint, Vredenburgh v. OpenAI OpCo, LLC, No. 3:26-cv-10527-AGT (N.D. Cal., filed September 16, 2026).
- Court docket on CourtListener: service on September 21, 2026, the October 13 response deadline and the case management schedule.
- 404 Media, “Inside Project Lily: The Humans Reading Your ChatGPT Chats” (September 14, 2026), the reporting the complaint relies on.
- OpenAI Help Center, “Data Usage for Consumer Services FAQ”; OpenAI Privacy Policy and “How your data is used to improve model performance”.
- OpenClassActions.com: Humans Are Reading Your ChatGPT Chats, New Class Action Claims — case summary, complaint PDF and status tracking.
This article is for informational purposes only and is not legal advice. Torts and Actions is not a law firm and is not affiliated with OpenAI or any party to the case. Everything described here is an allegation in a complaint: OpenAI has not yet responded in court, no class has been certified and no court has decided the merits. There is no settlement and nothing to claim. Anyone asking for a fee to “join” this case is not part of it.