When AI Screens Your Candidates, Who Is Legally Responsible for the Bias?

The assumption embedded in most AI-assisted hiring stacks is that automation neutralizes human bias. The Mobley v. Workday case suggests the opposite may be true — and that the legal bill will land on the employer's desk, not the vendor's.

Derek Mobley, a Black man over 40 with a mental health history, alleged that Workday's artificial intelligence (AI) screening tools systematically rejected his applications across multiple employers — making Workday, as a third-party vendor, potentially liable as an employment agency under U.S. civil rights law. The core finding, as reported by Personnel Today (2025-07-01), is that courts may treat AI screening vendors as legal actors in the hiring chain, not merely neutral technology providers. That single sentence rewrites the risk model most chief human resources officers (CHROs) built their AI procurement decisions on.

Why does delegating to AI not transfer the liability?

The answer lies in how employment discrimination law was designed. Anti-discrimination statutes attach liability to decisions, not decision-makers in the narrow sense. When a human recruiter makes a biased call, the employer is responsible. When an algorithm does, courts are now asking: who set the parameters, who chose the vendor, and who failed to audit the outputs? The employer is present at every one of those inflection points. The vendor, meanwhile, operates on scale — the same model rejecting Mobley was processing candidates for dozens of companies simultaneously, which is precisely what made the agency-theory argument compelling to the court.

Personnel Today (2025-07-01) notes that the case has ramifications for HR leaders globally, not just in the United States. In jurisdictions where data protection law (such as GDPR in the EU) already grants candidates the right to contest automated decisions, the legal exposure is arguably higher — because the regulatory framework is already explicit about human oversight requirements.

What does meaningful human oversight of AI hiring actually require?

This is where most organizations have a dangerous gap between policy and practice. "A human reviews all final decisions" sounds like a control. It is not a control if that human has no visibility into why the AI ranked candidates as it did, and no realistic time to question it. A reviewer rubber-stamping AI output is not oversight — it is liability theater.

Research from MIT Sloan Management Review (2025) adds a structural dimension to this problem: algorithmic tools can silently narrow an organization's decision-making by suppressing the value of human expertise. The researchers found that these systems tend to favor familiar patterns, which in hiring means they systematically underweight candidates who deviate from the historical profile of successful hires — compounding existing demographic imbalances, not correcting them. If your "diverse talent pipeline" strategy runs through a tool trained on your past hires, you have built a feedback loop, not a fix.

The McKinsey perspective on AI agents in the workforce frames this as a performance management problem: McKinsey & Company (2025) argues that AI systems require the same accountability infrastructure as human workers — defined outputs, monitored performance, and clear escalation paths when something goes wrong. Applied to recruiting AI, that means regular audits of pass-through rates by demographic group, defined thresholds that trigger human review, and documented rationale for every candidate rejection that an AI system initiates.

The inconvenient implication is that genuine human oversight costs more than the efficiency gains most vendors promised. Running a demographic audit on your ATS (applicant tracking system) outputs quarterly, training hiring managers to interrogate — not just accept — ranked candidate lists, and maintaining documentation sufficient to defend individual rejections in litigation: none of that is free.

The strategic decision you are actually facing

The Mobley v. Workday case does not mean AI has no place in recruitment. It means the compliance model built around "the algorithm decided" is no longer defensible. Organizations that treat AI screening as a liability-neutral cost-reduction tool are accumulating legal and reputational exposure with every hiring cycle.

The real question for the next board conversation is not "which AI hiring tool should we use?" It is "what is our audit and override infrastructure, and who owns it?" If your answer relies on a vendor SLA rather than an internal governance structure, you have your answer — and so, eventually, will a plaintiff's attorney.