# AI Contractors Shouldn’t Use AI to Evaluate AI Model, Says AI Company > Source: > Published: 2026-09-22 21:15:07+00:00 OpenAI has found itself in a “leopards ate my face” situation, except in this case, it’s a leopard of its own design. According to a new [report](https://www.404media.co/people-training-openais-ai-fired-for-using-ai-to-train-the-ai/) from 404 Media, the frontier AI lab had to fire a handful of its contractors who were tasked with reviewing ChatGPT outputs to ensure their accuracy because those workers found it easier to complete their tasks by just feeding them into an AI. Per the report, OpenAI employs thousands of contractors who it tasks with reviewing the prompts that people feed to ChatGPT and the outputs that the model spits back out to them. (Reportedly, those interactions can [contain sensitive personal information shared by users](https://www.404media.co/inside-project-lily-the-humans-reading-your-chatgpt-chats/), but that’s a whole other can of worms.) The contractors are tasked with taking a closer look at those interactions and rating them with the goal of making sure that ChatGPT is not only providing correct information, but also to keep the chatbot from getting too sycophantic in its responses. It’s repetitive, menial work that probably doesn’t pay all that well, which makes it the ideal type of task to pass off to AI. But that’s one of the things that OpenAI absolutely does not want its contractors doing, according to documents obtained by 404 Media. “Do not use AI detection tools, or AI yourself,” one document reportedly told the workers. “Do not use GPTZero or any other AI detection tool. They are not reliable. Reviewers may not use AI either, including Grammarly and AI translation, to review, write feedback, or write comments.” One contractor even [told 404 Media](https://www.404media.co/people-training-openais-ai-fired-for-using-ai-to-train-the-ai/) that using AI tools is “pretty much the one thing that will get you kicked off ASAP.” On one hand, it makes sense why OpenAI wouldn’t want to leave this particular task up to AI, as the AI-to-AI feedback loop could potentially contribute to [model collapse](https://www.ibm.com/think/topics/model-collapse)—a phenomenon that can occur when a model starts to be fed synthetic, AI-generated data instead of human-created content. A [2024 study published in Nature](https://www.nature.com/articles/s41586-024-07566-y) found that “indiscriminate use of model-generated content in training causes irreversible defects in the resulting models,” which is the kinda thing that you probably want to avoid if you’d like for your model to remain functional. On the other hand, Sam Altman has said repeatedly that [AI will take jobs](https://finance.yahoo.com/news/openai-ceo-sam-altman-warns-130002539.html) and that he worries that his job “[is to help people destroy jobs](https://www.instagram.com/reel/DYuUSRozUPO/).” So it’s a little funny to see his company see people surrender their jobs to AI and go, “Wait, no, not like that!” Gizmodo reached out to OpenAI for comment regarding the firings but have not yet received a response.