OpenAI Contractors Fired Over AI-Assisted Training

OpenAI’s human-feedback pipeline is exposing the labor contradiction at the center of AI: workers are expected to police automation while being punished for using it to meet demanding quotas.

• –AI-generated feedback can contaminate training data and contribute to model collapse.
• –The incident raises questions about how consistently contractors can perform detailed evaluation work under time and pay pressure.
• –Developers relying on model evaluations should track data provenance and distinguish human judgments from synthetic ones.
• –OpenAI’s strict policy is understandable technically, but the reported labor conditions make enforcement look economically motivated.
• –The broader industry may need auditable human-in-the-loop standards for training and evaluation datasets.