Stanford study finds AI hiring tools bias against Black and Asian candidates

A Stanford research team analysed a dataset of about 4 million job applications submitted to 156 employers across 11 sectors. The study revealed that AI‑driven screening tools, chiefly the Pymetrics platform, recommended Black applicants at a rate 26 % lower and Asian applicants 15 % lower than the most‑favoured group, meeting the EEOC’s “adverse impact” standard of less than 80 % of the top recommendation rate. The researchers estimate that eliminating the bias would allow roughly 40,000 additional applications from these groups to progress to the next hiring stage.

The bias stems from a “single‑vendor” market where a dominant AI model is deployed by many firms, causing systemic exclusion of certain racial groups across the job market. The findings arrive amid growing legal challenges, such as a U.S. district court case against Workday’s AI recruiting system and the recent Colorado AI Act that mandates reasonable care to prevent discrimination in AI hiring tools. The study underscores the need for transparent, evidence‑based AI policy and independent oversight of algorithmic hiring practices.