Mercor, a data-labeling startup that helps major tech companies train AI models, has sparked backlash after ending a large project for Meta called Musen and offering affected contractors a new role on a similar project, Nova, at a lower pay rate. Workers on Musen, which at its peak involved over 5,000 contractors, were informed the project was being terminated due to "project scope changes," according to an email seen by Business Insider. Soon after, they were invited to join Nova, which pays $16 an hour—$5 less than the Musen rate. The news came as a shock to many, as contractors had been told just weeks earlier that Musen would continue through the end of the year. One worker said the project had been repeatedly paused and restarted to accommodate different time zones, but the client, Meta, had consistently expressed satisfaction and promised extensions. “They kept stressing how happy the client was and how it got extended until the end of the year,” the worker said. “So to then do this major switch up before the holidays took everyone by surprise.” Mercor, founded by three Thiel Fellows and recently valued at $10 billion, is part of a growing industry of human-powered data labeling firms that are essential to training AI models. These companies employ thousands of contractors to tag and categorize text, images, and video, helping improve the accuracy and safety of AI systems. In the wake of Meta’s $14.3 billion investment in Scale AI and the hiring of Scale’s CEO Alexandr Wang, Mercor and other firms have gained new business from clients shifting away from Scale. Despite its claims of high pay, Mercor’s average rate of $95 an hour, as cited by CEO Brendan Foody in a September podcast, contrasts sharply with the $16 rate offered for Nova. Several contractors said they had previously earned up to $60 an hour on other Mercor projects. One worker who transitioned to Nova said the tasks were nearly identical but at a significantly reduced wage. “It sounds like most of us are in the same boat,” they said. “We wanted to boycott this but are not in a financial place to do so. We needed to have the guaranteed income, even if it’s demoralizing.” A Mercor spokesperson denied some details in the reports but declined to clarify what was inaccurate. The company reiterated that all roles are temporary and project-based, and that it provides transparency through role descriptions and onboarding materials. Meta declined to comment. The move follows broader layoffs and restructuring in the data labeling sector. Scale AI recently cut a team focused on generalist labeling, and Elon Musk’s xAI eliminated over 500 data labelers in September, shifting focus to higher-paid specialist roles. As AI development accelerates, the pressure on labeling firms to cut costs while maintaining quality is intensifying, raising concerns about worker treatment and long-term sustainability in the industry.