AI and work
A US court has ruled Workday's AI bias testing privileged because its lawyers curated the data
In a discovery order filed on 29 May 2026 in Mobley v Workday, the Northern District of California refused to compel Workday's bias-testing data, holding it covered by attorney-client privilege. It also refused the applicant data of Workday's customers, while ordering EEO-1 and OFCCP filings produced.

The case reported everywhere as "the AI hiring bias case" produced a ruling in May 2026 that almost nobody has reported, and its consequences reach well beyond the litigation.
On 29 May 2026, in Mobley v Workday, Inc., Case No. 23-cv-00770-RFL (LB) in the United States District Court for the Northern District of California, San Francisco Division, the court filed a discovery order (document 340) resolving three disputes. The order records that the plaintiffs are suing Workday "for utilizing an AI screening system that is more likely to deny applicants who are African American, suffer from disabilities, or are over forty years old", pleaded in a third amended complaint.
The court denied the motion to compel Workday's bias-testing data. Its stated reason is the important part: "Attorney-client privilege applies to the bias-testing data because Workday's attorneys curated the underlying data and used the results in providing legal advice." The order applies the eight part privilege test from United States v Ruehle, 583 F.3d 600 at 607 (9th Cir. 2009). Writing on 2 June 2026, Gerald L. Maatman Jr of Duane Morris, which published the order, summarised the holding as reinforcing "that bias-testing data can be shielded from production under attorney-client privilege when an employer's attorneys curate the underlying data and conduct bias-testing for the purpose of providing legal advice, as opposed to a business or regulatory compliance purpose", and noted the court had rejected the argument that referring publicly to the existence of bias testing waived privilege. His advice to clients was explicit: structure bias testing under the direction of counsel to preserve the privilege.
That is the finding with the longest reach. Algorithmic bias audits are widely promoted, and in some jurisdictions mandated, as the mechanism by which automated hiring tools are kept honest. This order indicates that where such testing is commissioned and curated by lawyers for the purpose of legal advice, and not submitted to a regulator, the results may never be discoverable by the people the tool screened out. The order draws the distinction along the purpose of the testing, not its content.
The court also denied the motion to compel the applicant data held on Workday's servers for its customers, finding the plaintiffs had not shown Workday has control of that data under Rule 34. Duane Morris reports the court rejected the argument that a master subscription agreement permitting production under court order amounted to control, because Workday lacked a legal right to obtain customer data on demand. In the one respect where the plaintiffs succeeded, the court ordered production of Workday's own EEO-1 and Office of Federal Contract Compliance Programs documents, holding them "relevant to Workday's knowledge of potential demographic disparities when utilizing the AI tools", a relevance sharpened by Workday using the same tools it sells. A separate application to deanonymise applicant data was terminated as moot.
The underlying liability theory was settled earlier. On 12 July 2024, District Judge Rita F. Lin denied Workday's motion to dismiss claims that it could be liable as an agent of the employers using its tools under Title VII, the Age Discrimination in Employment Act and the Americans with Disabilities Act, reasoning that an employer cannot escape liability by delegating traditional functions such as hiring to a third party, and declining to distinguish between algorithmic and human decision makers on the ground that doing so would gut anti-discrimination laws in the modern era. The Equal Employment Opportunity Commission had filed an amicus brief on 9 April 2024 arguing that screening and referral are functions classically associated with employment agencies, and that Workday's tools perform them by more sophisticated means.
The order also records what the tools are: Workday Recruiting, the Candidate Skills Match algorithmic feature, and, following Workday's 2024 acquisition of HiredScore, the Spotlight and Fetch products.
On scale, Forbes reported on 29 May 2026 that roughly 14,000 people had opted into the collective by the deadline of 7 March 2026, that Workday declined to confirm the number, and that the certification order dates from 16 May 2025. Workday's position, as quoted there, is that its "AI recruiting tools don't make hiring decisions and are designed with human oversight at their core". That is the company's characterisation of its own product.
What is unresolved: secondary accounts disagree about the dates of the 2026 motion to dismiss rulings, and this article does not rely on them. Nor is it known whether the privilege holding will survive review, or how many of the bias audits now being run across the hiring software industry are structured, as Duane Morris now advises, to be undiscoverable.
Sources
Every factual claim above rests on the 5 published sources below. They are listed so you can check the reporting rather than take it on trust.
- United States District Court for the Northern District of CaliforniaDiscovery Order, Mobley v Workday, Inc., Case No. 23-cv-00770-RFL (LB), ECF No. 340
- Duane Morris Class Action Defense BlogCalifornia Federal Court Clarifies Limits On AI Bias Testing And Applicant Data Disclosure In Mobley v. Workday
- Duane Morris Class Action Defense BlogCalifornia Federal Court Denies Motion To Dismiss Artificial Intelligence Employment Discrimination Lawsuit
- Duane Morris Class Action Defense BlogEEOC Weighs In On Novel Artificial Intelligence Suit Alleging Discriminatory Hiring Practices
- ForbesA Federal Judge, A 1967 Law, And A Billion Rejected Job Applications


