Mobley v. Workday tests who's liable for AI hiring bias
TL;DR
- A federal judge in San Francisco ruled June 22 that Derek Mobley's discrimination case against Workday can proceed after more than 100 rejections.
- Workday says its software does not make hiring decisions; it evaluates qualifications and the employer chooses the final hire.
- The European Parliament voted to postpone the EU AI Act's obligations for high-risk hiring systems to December 2027.
A federal judge in San Francisco let Derek Mobley's discrimination case against Workday proceed last month, and the reason it is worth attention has less to do with one applicant than with how his rejections happened. Mobley applied for more than 100 jobs through employers that screen applicants using Workday's software. Every one rejected him. The case, laid out in a TechPolicy.Press essay by Samonne Montgomery, now forces a court to answer a question the industry has quietly avoided: when an AI system filters out a candidate in ways that may violate California civil-rights law and the Americans with Disabilities Act, whose decision was it?
Workday's answer, per the piece, is that its software does not make hiring decisions. It evaluates qualifications, and the employer chooses. Montgomery's argument is that this framing does real work. When the outcome is convenient, the system looks autonomous. When the outcome is challenged, a human made the final call. Companies get to pick whichever version shields them, and the applicant is left with no identifiable decision-maker to appeal to. She borrows Madeleine Clare Elish's phrase 'moral crumple zone' to describe the human reviewer positioned to absorb legal blame for a design they cannot override.
The stakes for anyone deploying these tools are getting concrete. The Mobley suit names 'proxies for disability such as gaps in work history' as the mechanism, which is exactly the kind of surface-neutral input that vendor documentation tends to describe as merit-based. Meanwhile the regulatory picture is drifting the other way. The European Parliament voted to postpone the EU AI Act's obligations for stand-alone high-risk hiring systems to December 2027, and on June 25 Rep. Nathaniel Moran introduced an AI Incident Reporting Act aimed at frontier-model developers rather than hiring vendors. The gap between deployment and enforceable duty is widening.
The honest caveat is that Montgomery is arguing a thesis, and the court has only let the case proceed. Workday has not been found liable. What the piece does not resolve is how a court should actually apportion responsibility between vendor and employer, or what an auditable version of 'the employer chose' would even look like when the shortlist arrives pre-filtered. Those are the questions the Mobley discovery process and the next round of city audit rules will start to answer, and any HR leader relying on a screening vendor should be reading their contracts with that in mind.
Shared on Bluesky by 2 AI experts
-
“[I]f software cannot bear responsibility, the responsibility does not evaporate. It stays with the people and institutions that designed the screen, set its thresholds, supplied its data, configured it, bought it, and b…
View on Bluesky →
Originally reported by techpolicy.press
Read the original article →Original headline: Assigning Responsibility When AI Discriminates Against Job Applicants | TechPolicy.Press