Researchers examined 3.4 million applications processed by a single hiring vendor across 156 firms. Over 90 % of U.S. employers use such algorithms, creating an algorithmic monoculture that funnels candidates through identical AI screens. The analysis reveals stark disparities: 30% of Black applicants and 14.7 % of Asian applicants face adverse outcomes under Title VII standards.

Position‑by‑position testing uncovers bias hidden in aggregate data. Prior studies reported minimal impact because they pooled all roles, diluting outlier positions where discrimination spikes. Disaggregated results show 25,000 + Asian applications would have been recommended if each role selected candidates at the highest observed rate. This granular view aligns with legal requirements that assess impact per job.

Systemic rejection rates far exceed a baseline of independent decisions, with 10 % of four‑application seekers rejected across every posting. The excess homogeneity signals that shared reliance on one vendor amplifies correlated failures. Authors call for regulators to audit per‑position impact, monitor vendor concentration, and grant independent researchers data access to curb bias.