ChatGPT users file class action over Project Lily chat reviews

OpenAI has been paying hundreds of contractors more than $50 an hour to read, summarize, and score real ChatGPT conversations as part of an internal program codenamed Project Lily, a 404 Media investigation revealed this week, prompting a proposed class action lawsuit filed on Wednesday by two ChatGPT users who allege the company misled them about the privacy of their conversations.404media+1

Inside Project Lily

The investigation, published by 404 Media journalist Joseph Cox on September 14 and based on leaked internal documents and real user prompts, describes a structured review pipeline in which contractors read a user's prompt, summarize the user's intent, compare four ChatGPT-generated responses, and score them on a one-to-seven scale. The reviewers are not checking for factual accuracy. Instead, leaked training guides instruct them to penalize "AI-speak," flag responses where the chatbot claims to be human, and punish overly sycophantic answers — a behavioral pattern OpenAI has linked to real-world harm in separate litigation.aiweekly+2

OpenAI strips usernames and runs prompts through an automated privacy filter before they reach reviewers. But the company acknowledged to 404 Media that the filter is imperfect. Contractors reported seeing a "user memories summary" displayed above the prompts they reviewed, which could expose a person's profession, general location, or personal background drawn from prior conversations. "I don't think they would imagine some contractor somewhere is analyzing the conversations," one worker told 404 Media, referring to ChatGPT's more than 900 million users.404media+2

Legal Fallout

The revelations have already triggered legal action. According to Law360, a pair of ChatGPT users filed a proposed class action against OpenAI on Wednesday, alleging the company misled them about how private their conversations were. The lawsuit adds to a growing docket of litigation facing OpenAI, which is already defending wrongful death suits, copyright claims, and a separate California privacy case over data shared with third parties through embedded tracking tools.mediapost+2

In Europe, the disclosure compounds existing regulatory pressure. Italy's data protection authority has already fined OpenAI 15 million euros, largely for processing user data without adequate legal basis, and the Court of Justice ruling in EDPS v. SRB established that a company's duty to inform users applies at the point of data collection.thenextweb+1

How to Opt Out

Free, Plus, and Pro ChatGPT accounts have the "Improve the model for everyone" setting enabled by default; Enterprise, Business, and Education tiers do not. Users who want to prevent future conversations from reaching reviewers must manually disable the toggle under Settings and Data Controls — but opting out is not retroactive, meaning older conversations may still be used.pcmag+1