Employees Used AI While Training AI.. OpenAI Takes Action Against Them

OpenAI relies on contract workers to help improve its AI models by reviewing real ChatGPT prompts and conversations, assessing generated responses and identifying what users were asking. According to reports published in September 2026, several contractors were removed from projects after they were found using AI tools to complete tasks that were specifically intended to provide human judgment. Internal documents reportedly indicate that some OpenAI-related projects involve thousands of contractors.

The restrictions are aimed at preserving genuine human feedback, which plays an important role in evaluating and improving AI responses. Researchers have also warned about “model collapse” when AI-generated data is repeatedly used to train subsequent models. Studies indicate that excessive reliance on synthetic data can cause models to lose information and diversity over successive training cycles, potentially affecting the quality and accuracy of their outputs.