How AI labs use Mercor to get the data companies won’t share (techcrunch.com)

At TechCrunch Disrupt 2025 Mercor CEO Brendan Foody laid out how his startup connects former senior employees from banks, law firms and consultancies with AI labs (customers include OpenAI, Anthropic and Meta) to generate the high-value, domain-specific training materials those institutions won’t share. Mercor pays contractors—many still employed at their old firms—up to $200/hour to fill forms and write reports, claims tens of thousands of contractors, disburses roughly $1.5M daily, has grown to ~$500M ARR and recently raised at a $10B valuation. The company positions this expert-sourced marketplace as a faster, cheaper alternative to negotiating guarded corporate data deals. For the AI/ML community this model matters because it produces high-fidelity, workflow-level supervision ideal for fine-tuning and agent training in specialties like law, finance and medicine—areas where generic web data underperforms. The approach accelerates automation of complex tasks but raises technical and governance risks: ambiguous data provenance, potential IP or trade-secret leakage, compliance and auditability challenges, and concentration risk (few labs account for most revenue). It also signals a shift in data supply from low-skill labeling toward expert knowledge work, reshaping incentives, dataset quality, and the ethics/legal debates around what employees can monetize versus what companies can protect.