Identity Verification Failures in AI Marketplaces

This is the core scaling tradeoff in human data marketplaces, the faster a platform tries to onboard thousands of workers for sensitive AI work, the easier it becomes for bad actors, fake identities, and weak internal controls to slip through. Mercor’s model depends on AI interviews and rapid matching of a very large expert pool, which is exactly what makes identity checks, payment controls, and longitudinal worker history a product requirement, not a back office function.1 2 3 4

• Mercor was built to vet experts at high speed, using AI video interviews and tests across a large candidate base, then place them into hourly AI training work for labs like OpenAI and Meta. That workflow is efficient, but it also means a single weak link in screening or monitoring can propagate across many projects quickly.1 2
• The broader market is already splitting on this issue. Prolific emphasizes years of participant history, repeated verification, profiling, fraud detection, and a controlled supply base that rarely requires outside recruiting. That produces slower, more curated growth, but it directly addresses the impersonation and vetting problems that appear when expert labor is sourced at scale.3
• Comparable vendors show why this matters economically. Invisible and Mercor both sell human judgment into RLHF and evaluation workflows where customers are paying for trust as much as labor. If customers worry that the person labeling a medical or safety task is not who they claim to be, the platform is no longer just a staffing marketplace problem, it becomes a model quality and compliance problem.2 3 4

Going forward, the winning platforms in AI labor will look less like open marketplaces and more like controlled identity networks. The market is moving toward persistent worker records, deeper screening, tighter payment controls, and audit trails that let labs prove who did the work, on what basis, and with what quality history.3 4