OpenAI's contractors fired for using AI to do the work they were hired to do

OpenAI relies on thousands of contract workers to read real ChatGPT user prompts, rate the chatbot's responses, and provide the human judgment that supposedly keeps the model grounded. Some of those workers are using AI to do the job. OpenAI's subcontractors are firing them for it.

404 Media spoke with three contractors working on OpenAI projects and obtained internal documents describing the rules. One document, aimed at contractors who review other contractors' work, instructs them not to use AI themselves: "Do not use AI detection tools, or AI yourself. 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."

The same document tells reviewers not to explain what tipped them off. "Do not tell evaluators why you suspect AI. It is easier for them to hide if they know what you look for. Judge the overall pattern, not one clue."

Reviewers are told to watch for repetitive words, AI-style punctuation — including overuse of the em dash — and contractors finishing their work unusually fast. In Slack channels where workers ask each other for advice, one contractor said, people frequently post a sample with the question "Is this AI?" The answer, they said, is usually yes.

One contractor shared what they said was their termination letter, which cited problems with the "authenticity" of their work. "I'm not a bad person or worker. I just needed a little boost and turned to AI to help me which eventually led to my downfall," the contractor told 404 Media. "I felt no joy in the work or that I was contributing to society in any way."

A second contractor said people using AI is common and the consequences are swift: "In a group of thousands there are tons that have been caught." They called it "pretty much the one thing that will get you kicked off ASAP."

Two of the three contractors work for Mercor, an AI-training company that hires workers to review ChatGPT-related material. A Mercor spokesperson said in a statement that the company's contracts "strictly prohibit the use of LLMs to complete projects" and that when the company confirms a worker has used AI, "we immediately remove them from the project." The spokesperson said Mercor invests heavily in tools and systems to detect misuse.

The work these contractors do is meant to counteract a problem researchers call "model collapse" — the degradation that occurs when AI models are trained on AI-generated text rather than human writing. The people hired to keep that from happening are, in some cases, feeding AI-generated text back into the system themselves.

A fourth contractor, who has worked on training models for various AI companies, told 404 Media they deliberately choose the worst responses when rating prompts, in an attempt to sabotage the training. "I either pay zero attention to the results and choose randomly or purposely choose the worst output," they said. "I'm not sure how much of a difference it actually makes since there are hundreds of other people also rating prompt results, but it does feel like I'm getting paid to make AI worse."

OpenAI declined to comment on the firings. According to one internal document, the projects can involve more than ten thousand contractors.