Contractors Read Real ChatGPT Chats, A New Lawsuit Says.
Vredenburgh v. OpenAI, filed September 16 in San Francisco federal court, claims users were told ChatGPT was private while outside reviewers scored their conversations.
A backlit keyboard. Photo: Colin, CC BY-SA 4.0, via Wikimedia Commons.
A lot of people talk to ChatGPT like nobody else is in the room. A class action filed this month says somebody was.
The case is Vredenburgh v. OpenAI OpCo, LLC, No. 3:26-cv-10527, filed September 16, 2026 in the U.S. District Court for the Northern District of California and assigned to Magistrate Judge Alex G. Tse, according to the federal docket. Two California users are the named plaintiffs.
What the complaint says happened
According to the complaint as summarized by Top Class Actions and Open Class Actions, OpenAI ran an internal program code named "Project Lily." Under it, contractors hired through a staffing firm called Crossing Hurdles were given real user prompts and conversations. They wrote summaries, highlighted passages, and scored model answers on a one to seven scale, and that work fed back into model development.
The plaintiffs say ChatGPT was presented as a private, one to one exchange. They argue this kind of review isn't listed among the categories of recipients in OpenAI's privacy policy, and that the only mention is a Help Center answer saying "trusted service providers" may see content to improve model performance.
What they're asking for
The proposed class is every U.S. ChatGPT user, free or paid, leaving out Enterprise, Business, Team, Edu and API customers. The eight claims include California's Unfair Competition Law, the Consumers Legal Remedies Act, the False Advertising Law, the California Consumer Privacy Act, intrusion upon seclusion and unjust enrichment. They want damages, restitution, an injunction and a jury trial.
Where it stands
Nothing's been decided. No class has been certified, and these are allegations, not findings. OpenAI's response is due October 13, 2026, and the coverage we reviewed didn't include a comment from the company.
Why this belongs here
People have poured medical questions, relationship problems and work secrets into ChatGPT. If a court finds that strangers were scoring those chats while users thought they were private, it won't matter how good the next model is. That's a trust problem, and it's the kind that sticks.
Sources: CourtListener docket, Top Class Actions, Open Class Actions.
← Back to ChatGPT Disaster Homepage