Stanford-led study finds over 25% of Black applicants' submissions directed to positions where AI hiring algorithm triggers federal discrimination scrutiny

Researchers at Stanford University, Chapman University, and Northeastern University analyzed over 4 million job applications from 3 million applicants across 156 employers screened by Pymetrics (a talent platform owned by Harver). The study found 10.62% of individual job positions showed adverse impact on Black applicants, with 25.87% of all applications from Black applicants going to positions where the algorithm produced what federal guidelines define as discriminatory outcomes.

The researchers argue Pymetrics measured bias incorrectly by pooling all applicants and outcomes across employers and positions, rather than analyzing each of the 1,746 individual positions separately—the method U.S. employment discrimination law (the Equal Employment Opportunity Commission's 'four-fifths rule') is designed to apply. This aggregation approach masked per-position adverse impact.

The study documents 'systemic rejection': Pymetrics stores applicant scores for up to 330 days and reuses them across employers, so applicants receive the same evaluation from multiple companies without knowing it. Among applicants who applied to 10 positions screened by Pymetrics, 4% were rejected from every single one—a statistically higher rate than independent hiring decisions would predict.