Lawsuit Claims Contractors Read Real ChatGPT Conversations

Two ChatGPT users in the United States have filed a proposed class action alleging that OpenAI let outside contractors read real conversations on the chatbot. Vredenburgh et al. v. OpenAI OpCo, LLC, No. 3:26-cv-10527, was filed on 16 September 2026 in the federal court for the Northern District of California. No class is certified and nothing below is proved.

What the complaint alleges

The complaint says OpenAI routed real prompts and conversations to contractors who read them, summarised what users wanted and scored the chatbot's answers to improve its models. The work ran under an internal programme code-named "Project Lily", which 404 Media described on 14 September 2026; the suit followed two days later.

Contractors were recruited through a staffing firm, Crossing Hurdles, for jobs advertised as "AI data reviewer" and "chatbot evaluator". A reviewer picked a task, read a user's prompt — often a full conversation — wrote a short summary of what the user seemed to want, then scored four ChatGPT responses from one to seven. That feedback fed model development.

The privacy filter and past chats

OpenAI runs conversations through an automated "Privacy Filter" before reviewers see them, but the complaint says the filter's own documentation calls it a redaction aid, not a guarantee. A reviewer's screen can also show a summary of the user's earlier ChatGPT activity, which may reveal a name or where the person lives.

The disclosure gap

The Privacy Policy lists eleven kinds of outside companies that receive personal data, but none that labels or evaluates data by hand, the complaint says. The model-training page never mentions a human reader, and the one page that does disclose human review limits it to flagged content checked by OpenAI's own team.

Claims and next steps

Eight claims are brought: violations of four California consumer and privacy laws — the Unfair Competition Law, Consumers Legal Remedies Act, False Advertising Law and Consumer Privacy Act — plus fraud by omission, intrusion upon seclusion, invasion of privacy under the state constitution and unjust enrichment. The suit seeks damages and an order barring OpenAI from sending chats to outside reviewers without explicit opt-in consent, and asks for the "Improve the model for everyone" setting to be off by default.

The proposed class covers U.S. ChatGPT Plus users, with California and paid-subscriber subclasses; Enterprise, Business, Team, Edu and API users are excluded. OpenAI was served on 21 September 2026 and must respond by 13 October; the first case management conference is set for 18 December 2026.