Stories
OpenAI Fires Contractors for Using AI to Train ChatGPT, Citing ‘Model Collapse’ Risk
M
Moneycontrol•25-09-2026, 10:04
OpenAI Fires Contractors for Using AI to Train ChatGPT, Citing ‘Model Collapse’ Risk
- •OpenAI reportedly terminated contract workers who used AI tools to complete tasks aimed at improving the company’s AI models, specifically evaluating ChatGPT conversations.
- •Contractors were responsible for reviewing ChatGPT prompts and responses to prevent the chatbot from becoming overly sycophantic, with some earning up to $50 per hour.
- •The use of AI tools by contractors violated project rules, as OpenAI prohibits feeding AI-generated content back into its training process due to the risk of ‘model collapse’, which can reduce model accuracy and diversity.
- •Internal guidance reportedly banned the use of AI detection services, Grammarly, and AI translation tools even for reviewing contractor submissions.
- •Mercor, a company providing experts to AI labs like OpenAI, confirmed that its agreements prohibit LLM use for project work, with confirmed misuse leading to immediate removal.
Key Questions Answered
Contractors were fired for using artificial intelligence tools to complete their assigned project work rather than carrying out the evaluation tasks themselves, which violated strict project rules.
Feeding synthetic output back into training can lead to "model collapse," a phenomenon where AI models become less accurate and diverse.
Internal guidance prohibited contractors from using AI detection services, AI translation services, and tools such as Grammarly while evaluating submissions.
