Summary
- Two users of ChatGPT from California have initiated a proposed class action lawsuit against OpenAI, claiming that the company allowed external contractors to read their conversations without appropriate notification.
- The lawsuit focuses on "Project Lily," an internal initiative where contractors, hired via a staffing agency, summarize user prompts and evaluate chatbot responses on a scale from 1 to 7.
- The court has set a deadline of October 13 for OpenAI to respond; the plaintiffs are requesting that the default setting for "Improve the model for everyone" be turned off and that explicit consent is required before any human accesses a chat.
OpenAI is currently facing legal action regarding the privacy of user conversations on ChatGPT.
A proposed class action lawsuit was filed this month in the U.S. District Court for the Northern District of California, alleging that OpenAI directed actual user chats to outside contractors without proper disclosure. The company received the lawsuit on September 2.
Myriad: Which company IPOs next? Click to make your prediction.The complaint targets an internal OpenAI project known as Project Lily, which was first reported by 404 Media on September 14. According to the lawsuit, contractors, employed through third-party staffing firms for roles like "AI data reviewer" and "chatbot evaluator," access real ChatGPT prompts and conversations, summarize user intents, and score four different model responses based on a 1-to-7 scale.
This method is a fundamental aspect of how AI systems like ChatGPT improve. Essentially, human reviewers assess the AI's answers, and those evaluations are used to refine the model, a process referred to as reinforcement learning from human feedback (RLHF). The lawsuit asserts that users were not adequately informed that their conversations could be accessed by humans instead of just machines.
OpenAI does employ an automated system to filter conversations before they are reviewed by humans. However, the complaint alleges that this system is not foolproof, sometimes allowing personal information to reach contractors.
404 Media discovered that reviewers utilize a dashboard featuring a "user memories summary" that can disclose personal details such as a user's general location, job, or aspects of their personal life, despite usernames being removed. OpenAI has indicated that these reviews aim to mitigate two specific behaviors: the chatbot appearing overly human-like and exhibiting sycophancy, where it tells users what they want to hear rather than what is accurate.
ChatGPT boasts over 900 million weekly users, many of whom use it as a private diary, therapist, or search tool. Users often share sensitive information, including health issues, relationship problems, tax inquiries, and legal matters. The lawsuit contends that while OpenAI's privacy policy mentions certain categories of external parties that may access user data, it does not specifically identify data annotation or human evaluation vendors as such.
The complaint outlines eight distinct legal claims, including breaches of California's Unfair Competition Law, the Consumer Privacy Act, and common law claims like intrusion upon seclusion, a tort that addresses unauthorized invasion of someone's private affairs.
The plaintiffs are seeking damages, restitution, and punitive damages. Additionally, they are requesting that OpenAI implement specific measures: requiring opt-in consent before any conversation is reviewed by an outside party, changing the default setting for the "Improve the model for everyone" feature to off, adding a clear warning within the chat interface, and potentially deleting any work products linked to reviewed conversations while retraining models that utilized this data.
OpenAI's response is expected by October 13.