The Quick Version
- Multiple recent studies have found that AI resume screening tools and large language models rank identical resumes lower when they carry names associated with Black applicants.
- Legal protections have not kept pace: only a handful of cities require bias audits of these tools, and federal enforcement guidance has weakened even as adoption spreads.
The Gatekeeper You Never See
Most job applicants today have no idea whether a human being ever looked at their resume before they got a rejection email, or no email at all. A majority of large employers now use some form of automated tool to screen resumes, rank candidates or analyze recorded video interviews before a hiring manager sees anything. These systems promise efficiency and objectivity. What they are actually delivering, according to a growing body of research, is discrimination at a scale and speed no individual biased recruiter could ever match.
What the Research Keeps Finding
Researchers testing large language models used in hiring pipelines have repeatedly found the same pattern: identical resumes, differing only by the name attached, get ranked and rated differently depending on whether that name is statistically associated with Black or white applicants. A widely cited University of Washington study found resumes with Black associated names were selected as the top candidate by AI screening tools at meaningfully lower rates than identical resumes with white associated names, across a range of commercially available systems. This is not a fringe result. It echoes decades of pre AI audit studies, most famously a 2004 field experiment that found resumes with white sounding names received far more callbacks than identical resumes with Black sounding names.
Old Bias, New Container
The uncomfortable truth is that AI hiring tools are trained on historical hiring data, and historical hiring data reflects a labor market that has never treated Black applicants fairly. A model trained to predict who gets hired, based on decades of who actually got hired, will faithfully learn and reproduce that pattern unless someone deliberately intervenes to correct it. Most companies deploying these tools have little visibility into how the underlying model actually weighs a candidate, because the vendors treat that logic as proprietary.
The Law Hasn’t Caught Up
Regulation has lagged badly behind adoption. New York City’s Local Law 144 requires employers using automated employment decision tools to commission independent bias audits and publish the results, but it is one of the only laws of its kind in the country, and enforcement has been inconsistent even there. At the federal level, the Equal Employment Opportunity Commission issued guidance clarifying that existing civil rights law applies to algorithmic hiring tools, but that guidance carries no new enforcement mechanism, and recent shifts in federal priorities have deprioritized investigating these tools specifically. Employers can and often do treat vendor assurances about fairness as sufficient due diligence, with no independent verification required.
What Job Seekers and Regulators Should Demand
Black job seekers deserve more than blind trust in a black box. At minimum, applicants should have a right to know when an automated tool is making or heavily influencing a hiring decision about them, and a right to request human review before a rejection is final. Cities and states should follow New York’s lead and require independent bias audits as a condition of using these tools at all, with real penalties for employers who skip them. And employers themselves, if they are serious about the equity commitments many of them still claim to hold, should be auditing their own vendors before regulators force them to, not after a lawsuit or a viral study makes the choice for them.
The resume screening algorithm does not introduce new bias into hiring so much as it launders old bias through a process that looks neutral because it is automated. Fixing that requires treating algorithmic hiring discrimination with the same seriousness as the human kind, because to the applicant who never gets a callback, the harm is exactly the same either way.
