Class Action Suit Alleges OpenAI Routed ChatGPT Conversations to Human Reviewers Without Clear Disclosure

A federal class action lawsuit, Vredenburgh v. OpenAI OpCo, LLC, was filed in the U.S. District Court for the Northern District of California on September 16, 2026. The complaint alleges that OpenAI routed real ChatGPT prompts and conversations to outside contractors for review, summarization, and scoring to improve its models, a process allegedly code-named “Project Lily.” The suit claims this practice was not disclosed in OpenAI’s Terms of Use, Privacy Policy, or model-training pages, constituting a violation of consumer privacy laws.

Allegations of Undisclosed Human Review

The complaint relies on reporting by 404 Media to describe a review pipeline where contractors, recruited via a staffing firm, access real user conversations through a dashboard. These reviewers are alleged to:

• Summarize user intent.
• Read four ChatGPT responses.
• Score each response on a scale of 1 to 7.
• Provide rationales for model development.

While OpenAI reportedly uses an automated “Privacy Filter” to redact personal information, the plaintiffs argue this is a redaction aid rather than a guarantee. The complaint states that reviewer instructions require escalating tasks containing personal information and that reviewers may see summaries of past user activity that could reveal identities.

Disclosure Gaps and Legal Claims

The lawsuit centers on a discrepancy between OpenAI’s public disclosures and its internal practices. The complaint notes that OpenAI’s Privacy Policy lists eleven types of third-party vendors receiving personal data, none of which are explicitly identified as data-labeling or human-evaluation vendors. While a nested Help Center FAQ mentions that “trusted service providers” may access content to improve model performance, the plaintiffs argue this disclosure is buried and insufficient compared to the explicit warnings OpenAI provides for other features, such as employer workspace access.

The plaintiffs propose a nationwide class of all U.S. ChatGPT users, with specific subclasses for California residents and paying subscribers (Plus/Pro). Enterprise, Business, Team, Edu, and API users are excluded. The suit pleads eight causes of action, including violations of the California Consumer Privacy Act (CCPA), unfair competition, and intrusion upon seclusion.

Requested Remedies and Next Steps

The plaintiffs seek damages, restitution, and punitive damages. They also request injunctive relief, asking the court to:

• Bar OpenAI from sending conversations to outside reviewers without separate opt-in consent.
• Require the “Improve the model for everyone” setting to be off by default.
• Mandate clear warnings in the chat interface regarding human review.
• Potentially require the deletion of reviewer work products and retraining of models built from this data.

OpenAI was served on September 21, 2026, with a response due October 13, 2026. No class has been certified, and no settlement has been reached.