The Evidence
LLM-based resume screening consistently disadvantages candidates with Black and female-associated names even when all other resume content is identical.
Over 25% of Black applicants were affected by racial disparity in AI hiring algorithms.
The EEOC settled its first AI hiring discrimination matter in 2023. The Mobley v. Workday litigation is among the first major class actions alleging algorithmic bias in screening tools. The legal risk is real and growing.
Only 27% of employees fully trust their employers to use AI responsibly in HR decisions. 59% believe AI is making workplace bias worse. Trust is not there, and the evidence says it should not be assumed.
Three Verification Checks
Does the input or output use role-related stereotypes ("he speaks with authority," "she was enthusiastic")? AI will carry these asymmetries through as if they were neutral observations.
Does the synthesis apply the same attributes across all candidates, or does it quietly change the standard between them? Check for consistency, not just content.
Does the recommendation match what was actually said in the source notes, or is it an interpolation the AI invented? Go back to the raw notes if anything seems off.
Reza's story
Reza uses AI to synthesize interview notes from three interviewers for a promotion decision. Before acting, Reza runs the three checks. The stereotype check catches it: one interviewer described a male candidate as "commanding" and a female candidate as "nice," and the AI synthesis carried that asymmetry straight through as if it were neutral. Reza returns to the raw notes, re-evaluates on actual demonstrated competencies, and makes a fairer call. The AI did not create the bias, the interviewer did, but the AI laundered it into clean-looking analysis, and only the verification habit caught it.
Apply It Today
The next time AI synthesizes anything that feeds a people decision, run the three checks before you act. Look specifically for language that judges personality differently across people.
Reflect