Study Finds 25.87% of Black Applicants' AI-Screened Submissions Trigger Discrimination Scrutiny
Updated
Updated · Fortune · May 26
Study Finds 25.87% of Black Applicants' AI-Screened Submissions Trigger Discrimination Scrutiny
2 articles · Updated · Fortune · May 26
25.87% of applications submitted by Black job seekers—nearly 40,000 filings—went to positions where Pymetrics screening outcomes met the EEOC threshold for adverse impact, according to a Stanford-, Chapman- and Northeastern-led study.
4 million applications across 1,746 positions at 156 employers showed the disparities emerged only when jobs were tested position by position; the researchers say Pymetrics’ aggregate method masked legally relevant bias. Asian applicants also saw 14.74% of submissions land in discriminatory-outcome roles.
4% of applicants who applied to 10 Pymetrics-screened jobs were rejected by all of them, evidence of what the authors call “systemic rejection” as scores are reused across employers for up to 330 days.
25 applications would be needed to push the risk of systemic shutout below 0.1%, versus 10 if employers were making independent decisions, underscoring how one vendor’s model can lock candidates out across multiple companies.
60% of the Fortune 100 and eight of the 10 largest U.S. federal agencies were using HireVue algorithms as of May 2023, the paper says, as regulators tighten oversight ahead of EU AI Act hiring-tool rules taking effect on Aug. 2.
AI promised to eliminate hiring bias but created a 'systemic rejection' machine. Who is legally responsible now?
As new laws take effect in 2026, will companies be forced to abandon their AI hiring gatekeepers?
Secretly blacklisted by one AI, are you now locked out of hundreds of jobs without even knowing it?