Workday's screening algorithm gets a certified collective action.
A U.S. district court in the Northern District of California expanded certification of the Mobley v. Workday collective action in April 2026, allowing job applicants age forty and over who were rejected by Workday's AI enabled screening to opt in. The plaintiffs allege that the model produced disparate outcomes on the basis of age, and that Workday's role as a common vendor allowed a single class to reach across many customer employers.
Workday has denied liability and argued that customer employers control the criteria. That defense is now being pressure tested in discovery. The parties are trading requests for training data provenance, tuning history, and the exact configuration each customer applied. Analysts covering the case have said the discovery volume is unusually large for an employment matter.
Reuters coverage →The plaintiff bar has a template. It is portable across every AI hiring tool, every AI underwriting tool, every AI collections tool that touches a protected class. The theory routes around the vendor by naming customers who deployed the tool with specific configuration.
Vendor indemnity does not answer a magistrate's discovery order. Only a per candidate, per decision record does. If your record does not exist, you become the discovery volume that pushes the settlement number. Every HR, credit, and insurance leader running an AI screen is now on notice.
- candidate.decision: input features presented to the model, in order, with the exact tuning
- model.version: build hash, training snapshot reference, deployment window
- policy.applied: allowed factors, blocked factors, jurisdictional overlay
- human.reviewer: reviewer identity, override state, comment
- export: EEOC ready and magistrate ready on a single query
HR, credit, or insurance model in production?
Thirty minutes on the per decision record you would need to defend it.
Book a working session →