A Stanford-led analysis of 4 million applications sent through the Pymetrics hiring platform uncovered systematic rejection of Black and Asian candidates. The study, the largest of its kind, examined data from 156 firms, most with revenues above $5 billion, and found that applicants needed to submit roughly 25 separate applications to secure a single forward recommendation across the labor market.

Researchers measured “adverse impact” – a federal benchmark where a group’s selection rate falls below four‑fifths of the highest – and identified it in one out of ten roles for Black applicants and one out of twenty for Asian applicants. Shared algorithmic models linked 42 employers, meaning a rejection at one company likely replicated across others using the same model across similar hiring cycles.

The findings raise alarm for firms that rely heavily on game‑based assessments such as Pymetrics or HireVue, as vendor‑specific quirks now propagate across an entire sector. With four percent of candidates rejected by all ten applications receiving identical scores, the study suggests that AI screening can amplify bias rather than dilute it. Companies may need to audit or diversify their hiring tech stacks immediately for talent acquisition teams.