OpenAI dismisses contractors who used AI on ChatGPT review work
OpenAI hired contractors to review ChatGPT prompts and improve responses with human input. Some contractors used AI tools despite instructions not to, and OpenAI fired many after discovering this, according to 404 Media. The company aims to prevent model collapse, where models trained on AI-generated text degrade.
OpenAI hired contract workers to inspect ChatGPT user prompts and provide human feedback that would help refine responses. Their rules prohibited using AI or AI-detection software, specifically mentioning GPTZero, Grammarly, and translation tools, for reviewing, writing feedback, or commenting.
Some workers still turned to AI, and many were fired after this was discovered, according to 404 Media. The report did not specify how many. OpenAI wants to avoid model collapse, where models trained on text produced by earlier AI systems can worsen. One contractor said the practice was common and led to swift removal; OpenAI declined to comment.
This could affect contract reviewers, whose roles may become more closely monitored and less secure, and AI developers, who may need stronger checks that human feedback is genuinely human. If AI-generated training material spreads, users may see slower quality improvements or degraded chatbot outputs. The episode may also influence how AI companies weigh automation against human labor in future data work.