OpenAI contractors hired to grade ChatGPT’s answers and improve its models have lost their jobs after getting caught using AI to do that work, according to Joseph Cox at 404 Media, who spoke with three contractors and obtained internal documents. The documents ban contractors from using AI of any kind, even Grammarly or translation tools, and tell the reviewers who audit other contractors to look for repeated words, heavy em dash use and work finished suspiciously fast, on projects that can involve more than 10,000 people. Two of the sources worked through Mercor, which said its contracts forbid LLM use and that it removes anyone it confirms broke the rule, while OpenAI declined to comment.
