# Lawsuit Accuses OpenAI of Allowing Contractors to Review ChatGPT Conversations

By Riza Dagoc

Canonical URL: https://www.tokenpost.com/news/regulation/23889
Published: 2026-09-24T20:19:19.000Z
Updated: 2026-09-24T20:19:19.000Z

Two California users have filed a proposed class action accusing OpenAI of allowing contractors to read, summarize and score complete ChatGPT conversations, raising questions about privacy expectations for consumer accounts.

The complaint was filed Sept. 16 in the U.S. District Court for the Northern District of California, San Francisco Division, as case No. 3:26-cv-10527. It alleges that hundreds of contractors recruited through staffing firms could review chats, summarize a user’s apparent objective and score four model responses from one to seven.

The plaintiffs also allege that reviewer dashboards may show summaries of earlier ChatGPT activity, including location-related information. The complaint says users were not clearly told that human reviewers could access their conversations through a program identified as Project Lily.

OpenAI’s consumer-data policy permits authorized personnel and trusted service providers to access user content for abuse investigations, support, legal matters or model improvement unless users opt out. The policy warns: “Please do not enter sensitive information that you would not want reviewed or used.”

The lawsuit asserts claims under California’s Unfair Competition Law, Consumer Legal Remedies Act, False Advertising Law and California Consumer Privacy Act. It also alleges violations of California’s constitutional privacy protections, fraudulent concealment, intrusion upon seclusion and unjust enrichment.

The requested remedies include requiring opt-in consent before human review, turning off “Improve the model for everyone” by default, adding an in-chat warning, awarding damages and possibly deleting or retraining models.

Users can disable “Improve the model for everyone,” use Temporary Chat, manage memory settings, delete conversations or submit a privacy request. Temporary Chats do not contribute to model improvement, but OpenAI may keep a copy for safety monitoring for as long as 30 days.

OpenAI also applies a Privacy Filter when model improvement is enabled to identify and mask personal information. Its privacy guidance states: “Once this setting is off, new conversations still appear in chat history but are not used to train ChatGPT.”

The case centers on the difference between a user’s expectation of a private chat and the human-feedback processes used to improve artificial intelligence systems. OpenAI’s policies distinguish consumer accounts from Business, Enterprise, Education and API accounts.