Handshake targets label auditing infrastructure
Handshake
The Cleanlab move shows Handshake is trying to own the hardest part of AI data work, which is proving that expert labels are reliable enough to train and evaluate frontier models. Handshake already had a cheap source of credentialed supply through university relationships. Adding label auditing and data quality tooling pushes it closer to a software layer that checks, improves, and defends output quality, not just a marketplace that fills seats.
-
Mercor and Invisible scaled by organizing large pools of expert labor, but both models still look labor heavy. Mercor reportedly reached $500M annualized revenue with about $365M in contractor payouts, while Invisible used an internal workflow platform and 3,000 plus workers to deliver RLHF. Handshake is adding a quality control layer on top of similar labor supply.
-
That matters because buyers are shifting from broad annotation volume toward narrower tasks where mistakes are expensive, like reasoning traces, safety evals, and domain reviews in math, law, and science. In this market, the winner is less the company with the most contractors and more the one that can show which raters were qualified, how outputs were checked, and why the final dataset is trustworthy.
-
Prolific is the clearest contrast. It built matching, screening, fraud detection, API integrations, and deep participant profiling into the product from day one, which lets customers self serve high quality human data quickly. Cleanlab gives Handshake a path toward that same kind of infrastructure posture, but starting from a university talent graph instead of an academic research network.
The next step is a market where human data vendors look more like measurement and compliance platforms wrapped around expert networks. Handshake is well positioned to sell verified expert supply today, then move upward into audit trails, benchmarking, and continuous model evaluation, where quality proof carries more value than raw contractor volume.