The Northern District of California's collective-certification order turned a single job applicant's discrimination complaint into what may become the most consequential AI-hiring lawsuit in the country. Merits-phase discovery is now active. Here is what plaintiffs' counsel now has to solve.
The Northern District of California's collective-certification order in Mobley v. Workday, Inc. (3:23-cv-00770) turned a single job applicant's discrimination complaint into what may become the most consequential AI-hiring lawsuit in the country. The opt-in period closed March 7, 2026. Merits-phase discovery is now active. What plaintiffs' counsel decides to demand from Workday over the next twelve months will shape the technical evidentiary standard for AI-based hiring-tool litigation for years to come.
Derek Mobley filed the original complaint in February 2023, alleging that Workday's algorithmic applicant-screening tools rejected him from more than a hundred positions at companies using Workday's platform, and that the pattern was driven in part by age, race, and disability variables the model had learned. Two years of pleadings, a motion-to-dismiss cycle, and one round of amended complaints followed. The district court's key rulings shaped the case's shape:
The case is being tried by Wiggins Childs Pantazis Fisher & Goldfarb LLC (Robert L. Wiggins Jr. named partner) and Winston Cooks LLC (Roderick Cooks) as counsel for plaintiffs. Sources: Civil Rights Litigation Clearinghouse, Wiggins Childs firm news, HR Dive, Akin Gump AI Regulation Tracker.
Certifying a collective is one thing. Proving disparate-impact discrimination in a merits-phase trial where the model in question is a proprietary black box owned by a third-party vendor is another. The technical burden of proof under Title VII / ADEA jurisprudence generally requires plaintiffs to (a) identify a specific employment practice or decision rule, (b) demonstrate statistical disparate impact on a protected class, and (c) show that the practice is not job-related or consistent with business necessity. Every leg of that test lands on discovery of Workday's model.
Three technical obstacles that will define the merits phase:
1. Model opacity. Workday's applicant-screening tool is not a single model but a system of ranked-decisioning components. Plaintiffs need production of the training-data specification, the model architecture, the version history, the feature-engineering rules, and the actual decisioning-output logs at customer-employer granularity. Workday will resist on trade-secret grounds. Judge Corley's discovery-management rulings over the next twelve months will shape whether the plaintiff class can see enough of the box to prove anything.
2. Vendor-liability structure. Workday is not the employer that rejected Mobley. Workday is the vendor that sold the tool that made the rejection. The theory of joint liability that plaintiffs are pursuing - that Workday's provision of a screening tool sufficiently controls the hiring decision to trigger direct ADEA liability - is not settled law. If it stands, it opens direct-liability exposure for every algorithmic-hiring vendor operating in the United States. If it falls, plaintiffs are pushed back to individual-employer defendants who can raise "we relied on the vendor" as a shield.
3. Rule 702 gatekeeping. Any expert methodology plaintiffs put forward to demonstrate disparate impact will face a Daubert challenge from a well-resourced Workday defense. The post-Kohls exclusion pattern (federal Rule 702 as amended in 2025) explicitly targets expert methodologies that rely on machine-learning inference to analyze machine-learning outputs. Plaintiffs' expert witnesses in the AI-discrimination space have to build their analytical pipelines out of pre-AI statistical primitives - Fisher F-test, Pearson chi-square, Kullback-Leibler divergence, Cohen's kappa on human labeler agreement - or face exclusion at the gatekeeping step. That is not a small constraint.
Based on the shape of the pleadings and the technical questions the case raises, plaintiffs' counsel will likely need production of some or all of the following from Workday:
Any expert methodology that ends up analyzing this production has to survive Rule 702. The path that works: classical statistical estimators applied to the discoverable output, capture of every measurement in a cryptographically-hashed evidence record produced independently of Workday, and independent-verifier declarations attesting to methodological independence from any audited vendor. The path that gets excluded: importing an ML model to analyze Workday's ML model's output.
Every industry that ended up regulated - pharmaceuticals, medical devices, aviation, financial services, automotive - walked a period in which the plaintiffs' bar took the first cases before the regulator was fully staffed and the rules were fully written. That work forced disclosure. It created discovery records that later became rulemaking citations. It named specific defendants and specific harms in ways that legislators could pick up and codify.
The AI arc runs faster than any of these because the plaintiffs' bar has already built the litigation infrastructure, the courts are already writing doctrine, and governments are already positioned to react. Mobley is one of the first two or three cases in the AI-hiring vertical that will produce the discovery record that later informs federal harmonization. A settlement or verdict in Mobley would move the AI-hiring space faster than any single piece of guidance from the Equal Employment Opportunity Commission has moved it in the past three years.
Bias-audit requirements are already in force in New York City (Local Law 144, effective July 2023) and, in progressively iterated form, in Illinois (Artificial Intelligence Video Interview Act) and California (proposed CCPA regulations on automated decisionmaking). Federal harmonization comes next. The specific shape of the harmonization - what audit methodology counts, what disclosure obligation triggers, what joint-liability structure applies to vendor-employer relationships - will be written with the Mobley discovery record on the table.
For attribution in a subsequent piece of this Case Spotlight series, comment has been requested from the following counsel and press representatives on the state of Mobley discovery and the questions it raises for the AI-hiring vendor market. Comments will be published verbatim where provided; declined-to-comment status will be noted.
If any comments arrive after publication, an addendum will be posted with attribution.
Mobley v. Workday is the case that will establish or reject the theory that an AI-hiring vendor can bear direct liability for the discrimination that its tool enables in the hiring decisions of its enterprise customers. The merits phase now determines both the specific outcome and the technical evidentiary standard for the next decade of AI-hiring litigation.
The methodology plaintiffs choose has to be Rule 702-defensible under the post-Kohls tightening. The discovery plaintiffs demand has to be scoped to withstand Workday's trade-secret objections. And the theory of liability plaintiffs plead has to land somewhere between "Workday just sold a tool" and "Workday is the employer" - a middle position that no federal court has yet written into black-letter law.
Every hearing in this case for the next twelve months matters. Whichever way it lands, the AI-hiring vertical will not look the same on the other side.