Two users sue OpenAI over contractor reading of chats

Decrypt reports that two California residents submitted a class action complaint to the U.S. District Court for the Northern District of California on September 2. The lawsuit alleges the company directed private user conversations to external contractors without sufficient disclosure or prior agreement.

The dispute focuses on Project Lily, an internal program first reported by 404 Media on September 14. Per the filing, contractors engaged through third-party staffing agencies examined actual user prompts and complete chat logs. These personnel assessed user intent and graded four unique model outputs based on a 1-to-7 metric. This evaluation procedure, referred to in the sector as reinforcement learning from human feedback, depends on human ratings to adjust model performance.

The complaint contends that OpenAI’s privacy policy enumerates specific data access types but does not explicitly name data annotation or human evaluation vendors. The plaintiffs claim this gap breaches California’s Unfair Competition Law, the California Consumer Privacy Act, and common-law torts including intrusion upon seclusion. The filing outlines eight distinct legal grievances against the corporation.

Plaintiffs are requesting damages, restitution, and punitive penalties. They are also applying for a preliminary injunction to compel OpenAI to disable the "Improve the model for everyone" setting by default. The court must mandate opt-in consent before any human examines a conversation, and the plaintiffs seek removal of work products linked to reviewed chats. OpenAI is obligated to reply to the submission by October 13.