Start-ups pay white-collar workers to train AI that may replace them

Mercor paid more than $2 million a day to contractors training artificial intelligence for companies including OpenAI, Anthropic, Meta and Google, even as the San Francisco startup’s own pitch centers on organizing human intelligence to power the AI economy. Founded in 2023, Mercor has built a business around highly paid experts in law, medicine, investment banking, consulting and STEM fields, with its site listing hourly rates that can reach about $180 for some medical specialists and $150 for legal experts.

The scale has made Mercor one of the clearest examples of a new labor bargain inside the AI boom: white-collar professionals are being paid to teach systems the tasks that give their own careers value. Reporting in January 2026 said the company had hired more than 30,000 contractors in 2025, a rapid expansion that helped turn expert labor into a high-volume data supply chain. Mercor’s $10 billion valuation after a $350 million Series C round led by Felicis in October 2025 showed investors were willing to pay for that model.

The work has drawn scrutiny because the people feeding these systems are often helping automate their own professions. Contractors and observers have described the arrangement as a strange new gig economy, and in some cases as intrusive or miserable, with some workers saying they had to sign nondisclosure agreements. The model is especially stark in fields that have long depended on specialized credentials and judgment, including hematology and oncology, legal practice, investment banking and management strategy.

That tension fits the broader labor warnings now coming from public institutions. On March 11, 2025, the U.S. Bureau of Labor Statistics said artificial intelligence is expected to primarily affect occupations whose core tasks can be replicated by generative AI in its current form. The International Labour Organization’s 2025 update went further, saying about one in four jobs globally could be transformed by generative AI, while also stressing that actual automation is likely to fall below the maximum exposure estimates.

The issue moved further into the mainstream on June 4, 2026, when The Journal aired an episode titled How AI Is Being Trained to Do Your Job, focusing on Mercor and the fast-growing contractor network behind it. Days after Mercor disclosed a March 31, 2026 cyberattack tied to a compromise of the open-source LiteLLM project, scrutiny of how these firms collect and store expert labor sharpened again. The central question is no longer whether AI will reach white-collar work, but how long workers will keep getting paid to train the systems that may eventually replace them.