OpenAI employs a vast network of contractors to review ChatGPT user prompts and data to provide a "human touch" to its models. However, reports from 404 Media indicate that multiple contractors have been fired for using AI to perform their work.
Internal documents and testimonies from three contractors reveal that workers are strictly prohibited from using AI tools—including Grammarly and AI translation services—to review, write, or comment on data. Reviewers are instructed to watch for "tell-tale signs" of AI use, such as repetitive wording, specific punctuation patterns, or unusually rapid completion of tasks.
One contractor, who provided a termination letter citing issues with "authenticity," stated that they turned to AI to assist with the workload. Two of the contractors spoke to 404 Media on the condition of anonymity.
The use of AI-generated content to train new models is linked to "model collapse," a phenomenon where models become progressively worse as they are trained on synthetic data.
Mercor, an AI-training company that provides contractors for such projects, stated that its contracts strictly prohibit the use of Large Language Models (LLMs). A spokesperson for the company confirmed that they invest in detection tools and immediately remove experts from projects if AI misuse is confirmed.
OpenAI declined to comment on the reports regarding its contractors being terminated for using AI.
Sources
- People Training OpenAI's AI Fired for Using AI to Train the AI （Hacker News Frontpage, 2026-09-22）