OpenAI faces privacy lawsuit and contractor dismissals over AI training

OpenAI is facing a proposed class-action lawsuit in Northern California regarding user privacy. The legal action alleges that the company allowed outside contractors to read and score real user conversations through an internal initiative known as ‘Project Lily’ without providing adequate disclosure or obtaining proper consent.

Simultaneously, reports have surfaced regarding the dismissal of numerous contractors involved in training OpenAI’s models. According to internal documents and worker accounts, these individuals were ‘offboarded’ or fired for using AI tools to complete the very tasks they were hired to perform, such as evaluating ChatGPT responses.

Companies like Mercor, which supply reviewers for these projects, have reportedly enforced strict rules against using large language models to ensure the authenticity of human feedback. This is intended to prevent ‘model collapse,’ a phenomenon where AI models deteriorate by being trained on synthetic, AI-generated data rather than genuine human input.