OpenAI Contractors Removed Over Unauthorized AI Use
The firm terminated contractors for using AI tools to assist in reviewing prompt and response quality.
Updated on Sept. 25, 2026 in Artificial Intelligence
Live Poll
Do you believe it is acceptable to use AI tools to complete professional work assignments?
OpenAI has removed multiple contractors from its projects after discovering they used artificial intelligence to perform assessment tasks. These contractors were responsible for reviewing user prompts and chatbot responses to ensure quality.
Why it matters
The company relies on human review to refine its models, making the use of unauthorized AI tools during these evaluations a direct threat to the integrity of the training data. The policy exists to ensure that human-in-the-loop assessments remain strictly human-driven.
Thousands of contractors are employed to review model outputs, but internal guidelines mandate that all tasks must be completed manually. The use of AI, Grammarly, translation software, and AI detection tools is explicitly forbidden.
The players
OpenAI
A research and deployment company focused on large-scale artificial intelligence models and human-in-the-loop feedback systems.
Mercor
An AI-training firm that manages and supplies contractors to review ChatGPT-related materials.
The details
Contractors are tasked with evaluating user prompts and chatbot responses to maintain quality standards for OpenAI. Internal company guidelines strictly prohibit the use of any assistive software, including AI-based platforms, for these assessments. This policy ensures that the review process remains a human-only feedback loop for model training.
Timeline
September 25, 2026: The reported removals were confirmed.
The Tech Race
Maintaining human-verified data is a critical bottleneck in the race to improve AI model reasoning and output quality. OpenAI's enforcement mirrors the broader industry struggle to prevent AI-generated noise from corrupting the very training sets intended to teach future models.
This enforcement action primarily impacts the labor practices of contractors hired through partners like Mercor. Users of ChatGPT will not notice immediate changes to the interface, though the policy change underscores the ongoing technical challenge of validating model accuracy.
The takeaway
The integrity of AI training data depends on the strict separation of human reviewers from the very tools they are tasked with evaluating. Observers should monitor whether other AI labs implement stricter automated detection to prevent contractors from using AI-assisted review tools.
Further reading
For broader context on the industry's approach to data integrity, explore our coverage of Artificial Intelligence.
Source note: This article includes information reported by Scottish Legal News.
Live Poll
Do you believe it is acceptable to use AI tools to complete professional work assignments?
