AI Hiring Discrimination: Who’s Liable When the Algorithm Decides?

The liability that never moved.

The through-line is simple: rules churn; liability doesn’t. Executive orders redirected federal enforcers, but no statute moved. Title VII’s disparate-impact framework remains codified at 42 U.S.C. § 2000e-2(k), and private plaintiffs proceed under Title VII, the Age Discrimination in Employment Act, and the Americans with Disabilities Act. Executive Order 14281, signed April 23, 2025, directed federal agencies to deprioritize disparate-impact enforcement, but it amended no statute and does not reach private suits or state law.

The federal retreat did not close the exposure; it redirected it. The EEOC closed disparate-impact-only charges by September 30, 2025 and issued right-to-sue letters by October 31, 2025 — moving those matters into federal court rather than out of existence. Lex Machina’s March 2026 report counted more than 20,000 federal employment-discrimination suits filed in 2025, the most in its dataset’s history.

Mobley v. Workday: the case that names the problem.

Mobley v. Workday, Inc. (N.D. Cal.) is the reference case for the question every HR leader is now asking — is our AI hiring tool our liability? Its answer, so far, is that the tool does not move the liability, and that the vendor can be pulled in alongside the employer rather than in place of it.

The chronology, as of July 12, 2026:

• July 12, 2024 — the court allowed discrimination claims to proceed against Workday as an agent of the employers using its tools. The agent theory is additive: it does not shift liability off the employers, who remain liable as principals.
• May 16, 2025 — the court preliminarily granted nationwide collective status under the ADEA, later extended (July 7, 2025) to Workday’s HiredScore features.
• March 6, 2026 — the court sustained the disparate-impact theory.
• May 29, 2026 — a discovery order routed requests for customer-employer data through the employers directly. Notice had gone out in January 2026 to a population Workday itself described as covering hundreds of millions of applications; roughly 14,000 opt-ins were reported by the March 2026 deadline — a figure Workday declined to confirm.
• July 2 and July 10, 2026 — Workday’s bid for interlocutory appeal was denied, and a Fourth Amended Complaint was filed.

The one-line consequence for an employer: the employers’ own files are now the discovery record. What you can show — the tools you used, what they produced, and what a human decided — is what answers the claim.

The pattern around it.

Mobley is the clearest statement of the theory, but it is not alone. Different vendors, different statutes, one constant — the theory of liability is portable and the tools are everywhere:

• EEOC v. iTutorGroup — a consent decree entered September 8, 2023 ($365,000) over recruiting software programmed to auto-reject older applicants.
• Baker v. CVS Health (D. Mass.) — a Massachusetts lie-detector-statute theory over HireVue/Affectiva AI video-interview analysis; the parties notified the court of settlement July 17, 2024.
• The ACLU’s FTC complaint over Aon’s assessments (filed late May 2024), and ACLU charges filed with the EEOC and the Colorado Civil Rights Division over an Intuit/HireVue promotion decision (announced March 19, 2025).
• Harper v. Sirius XM Radio (E.D. Mich., filed August 4, 2025) — Title VII race claims over an AI applicant-screening tool.
• Kistler v. Eightfold AI (filed January 20, 2026; removed to N.D. Cal. April 29, 2026) — the theory that AI screening scores are consumer reports under the FCRA.

What it means for employers.

The exposure is national and it attaches to the employer. That has four practical consequences. First, candidates increasingly must be told when AI factors into a covered decision — the specifics vary by jurisdiction. Second, adverse decisions must be explainable, which means a human review that is actually documented. Third, records must exist before a charge arrives, because the file is the defense. Fourth, the vendor must answer: written responses on training data, bias testing, and protected-class impact are the diligence record, because under this theory the vendor’s tool is the employer’s exposure.

None of that is a single document or a one-time project. It is a maintained posture — inventory, notice, vendor answers, testing, retention, and named ownership — kept current as the tools and the rules change. That is the work the TalentSight Intelligence library is built to support, and the free AI Hiring Exposure Snapshot is a four-minute read on where an organization’s exposure surface actually is.

The insurance signal.

Insurers moved before regulators did. ISO introduced generative-AI exclusion endorsements with January 2026 edition dates; at least one national carrier runs an “absolute” AI exclusion across management-liability lines; and employment-practices-liability underwriters have begun asking applicants whether they use AI hiring tools. For many organizations, the first hard question about AI in hiring will come at renewal — from the carrier — not from a regulator.

How to cite this article.

Abdullahi, Khullani M. “AI Hiring Discrimination: Who’s Liable When the Algorithm Decides?” Techné AI, 2026, https://techne.ai/insights/ai-hiring-discrimination-employer-liability.