A significant shake-up has hit the data-labeling sector: Mercor, a rapidly growing startup that supplies contract workers to major AI labs including Meta Platforms and OpenAI, abruptly ended a project known as “Musen,” affecting thousands of contractors. Many were then offered similar work on a new initiative called “Nova,” but at a reduced wage of US $16 per hour, about $5/hour less than what they had been earning on Musen.
At its peak, the Musen project reportedly involved over 5,000 contractors, all of whom were surprised by the sudden termination—especially since many had previously been told the project would run until the end of the year. The company provided limited comment, stating only that “transparency” is maintained around the temporary nature of roles, but critics say the sudden shift before the holiday period was demoralizing for workers.
Mercor, which claims to manage over 30,000 contractors and recently reached a valuation near US $10 billion, says it pays more than US $1.5 million daily to its workforce. Yet the pay-cut and project shift highlight the precarity of what the company calls a “new category of work” — human trainers now teaching machines how to think, only to find their own terms of employment can change instantly.
This episode underscores the wider instability within the data-labeling and AI-training industry: as AI firms seek scale and efficiency, contract labourers face volatile project terms, pay cuts and little job security. It raises questions not only about worker rights, but also about dependence on gig-style models for training the technology powering AI.
