The Algorithm Said No 100 Times. Now a Judge Says the Algorithm Can Be Sued.

A federal judge just told Workday it can't hide behind “we're only the software.” US District Judge Rita Lin signaled on June 16 she'll likely let sprawling discrimination claims against the HR-software giant move forward — including claims it violated California's Fair Employment and Housing Act thousands of times by screening out applicants at major companies nationwide.

The case — Mobley v. Workday — was brought by Derek Mobley, a Black, disabled man over 40 who says Workday's AI screening tools rejected him from more than 100 jobs. Workday's defense: it isn't the “employer,” so anti-discrimination law shouldn't touch it. The court's answer: a vendor whose tools materially influence who gets rejected can be treated as the employer's “agent” — and held directly liable.

The bias doesn't need a checkbox for race or age to do its damage. Years of experience signals age. Employment gaps infer disability or caregiving. School and institution names can proxy for race. The model learns the correlation and reproduces the outcome at scale — no human ever in the loop to catch it.

This is the flag that matters: when the machine becomes the gatekeeper, “it was just the tool” stops being a defense. HITL Score: 18/100 — human oversight at deployment 5/25 (auto-screening rejected 100+ applications with no human gate on the no-decision; the system was trusted to filter people out before anyone saw the résumé), ongoing monitoring 4/25 (no evidence of disparate-impact testing across age/race/disability; the bias entered through training data and “fit” criteria and went unmeasured at scale), incident response 5/25 (100+ rejections to one applicant produced no flag, no review, no correction loop — the pattern itself was invisible to the operator), accountability 4/25 (the vendor's posture was “we're not the employer, not our liability”; the court had to construct the accountability the deployment never built in).