404 Media reported that several contractors on OpenAI-related projects were fired or removed after using AI to rate chatbot responses, despite rules barring AI assistance. Internal documents cited by the outlet say some projects involved more than 10,000 contractors.
The cited instructions barred AI use—including Grammarly and AI translation—for reviewing, writing feedback or comments, and also prohibited AI-detection tools. In separate reporting, 404 Media described Project Lily, an OpenAI effort with hundreds of contractors who review real users’ prompts and conversations, which can contain personal information, and score ChatGPT’s replies. The outlet also cited the risk that further training on AI-generated text could gradually worsen models, a process known as “model collapse.”
Mercor, an AI-training firm connected to the work, said its contracts prohibit LLM use and that it immediately removes experts when it confirms AI was used to complete a task. OpenAI declined to comment on claims that contractors were fired for using AI.
