OpenAI wants you to use AI — but not to train its AI
OpenAI fired contractors for using AI to review AI. The reason: model collapse—and it's happening at scale.
Why it matters
OpenAI discovered many of its data-labeling contractors were using AI tools to complete human-review tasks meant to improve model quality. This undermines a core mitigation against model collapse—a documented phenomenon where AI trained on AI-generated text degrades. The incident reveals both the operational challenge of enforcing quality boundaries in contractor networks and the real cost of the collapse risk to frontier labs.
The key facts
14 to knowOpenAI hired contractors to read ChatGPT prompts and provide human feedback to improve responses
Many contractors used AI tools (including Grammarly, GPTZero, other AI detection tools) to complete the work despite explicit contractual prohibition
Contractors' working conditions explicitly stated: 'Do not use AI detection tools, or AI yourself… Reviewers may not use AI either, including Grammarly and AI translation'
OpenAI fired multiple contractors for the violation; exact number not disclosed by 404 Media
One anonymous contractor told 404 Media the practice was common: 'In a group of thousands, there are tons that have been caught'
OpenAI's stated concern: model collapse—AI models trained on AI-generated text perform worse than predecessors
OpenAI declined to comment on the scope or scale of the enforcement action
Report source: 404 Media (original reporting)
OpenAI contractors using AI to generate feedback despite explicit contract prohibitions
Contractor reports widespread practice: 'tons have been caught' in groups of thousands
Model collapse risk: AI trained on AI-generated text performs worse than predecessors
Contract terms explicitly forbade AI tools including Grammarly and GPTZero
OpenAI declined to comment on the number of contractors fired or prevalence
Incident reported by 404 Media; no independent verification of scale
Go to the source
Computerworldcomputerworld.com
Publisher excerpt: Here’s an interesting concept: an AI company that fires people for using AI. It sounds like a strange way to run a company but there’s a real reason behind it. OpenAI has been hiring contractors who have been tasked with reading ChatGPT users’ prompts to help improve responses by providing some…