Behind the privilege shield: Safeguarding AI bias-testing data in employment decisions

On May 28, 2026, a California federal judge held in Mobley v. Workday that artificial intelligence (AI) bias-testing data may be protected against discovery based on the attorney-client privilege shield. The ruling carries practical significance for employers navigating an expanding patchwork of regulations governing automated employment decision-making, particularly those seeking to validate their AI systems without sacrificing the privilege of their legal strategies.

Background

We recently examined the collective action against Workday in Mobley v. Workday, Inc., Case No. 23 -cv-00770-RFL (LB) (N.D. Cal. May 28, 2026). As a brief recap, Workday is an “AI platform for HR and finance.” It is used by over 11,000 entities around the world to “elevate their people, supercharge work and move their business forward forever.” Among the many services it provides, Workday offers AI tools that use advanced algorithmic methods to screen and ultimately accept or reject applicants for these entities. One such tool, known as "Candidate Skills Match," is an algorithmic product that determines how well an applicant's skills match a particular role. Workday has publicly stated that its "AI recruiting tools don't make hiring decisions and are designed with human oversight at their core."

Derek Mobley (Black, over age 40) alleges that, since 2017, he has applied to more than 100 jobs with companies that use Workday’s screening features for talent acquisition and hiring. According to Mobley, he was rejected every single time. Mobley initially filed an individual lawsuit in the US District Court for the Northern District of California against Workday, alleging race discrimination under Title VII and Section 1981, age discrimination under the Age Discrimination in Employment Act (ADEA) and disability discrimination under the Americans with Disabilities Act (ADA). In May of 2025, the court granted conditional collective action certification for the ADEA claims, and a motion for class certification on the remaining claims remains pending.

Workday, like many organizations, is subject to New York City Local Law 144 (NYC 144) and its implementing regulations, which went into effect on July 5, 2023. NYC 144 applies to employers and employment agencies that use an Automated Employment Decision Tool (AEDT) to screen a candidate or employee for an employment decision, regardless of where the entity is headquartered. The law prohibits the use of an AEDT unless the tool has been subject to a bias audit conducted within the prior year. Entities must use and assess historical data as part of a bias audit to ensure that the AEDT is performing as expected. A summary of the audit results (not the audit itself) must be made publicly available on the employer’s or employment agency’s website. Further, entities must comply with certain notice requirements to the candidate or employee. For further information on these requirements, see our analysis of the regulation.

Anti-bias testing and privilege

Importantly, Workday agreed to and had already produced the underlying historical data used as part of its bias audit to plaintiffs in the course of discovery. The only issue was the bias-testing data, which Workday claimed was irrelevant and/or protected by the attorney-client privilege. Plaintiffs moved to compel Workday to produce its bias-testing data. The court ordered Workday to produce a privilege log identifying the documents withheld from plaintiffs’ requests for production.

In support of its claim of privilege, Workday asserted that it conducted bias testing of various products and features through its lawyers for the purpose of obtaining legal guidance regarding these products, including guidance on compliance with the bias testing approach prescribed by Local Law 144. The products tested included Candidate Match, an algorithm designed to determine the extent to which an applicant’s skills match the role for which they applied, and Spotlight, a candidate review tool. Workday did not claim privilege over certain anti-bias testing of Spotlight where it engaged an external consultant to test the product using the impact ratios approach described in NYC Local Law 144. Plaintiffs, by contrast, asserted that the data was not privileged because Workday’s counsel were acting in a business capacity, and notably, certain bias testing data was submitted to an external auditor. The parties agreed that the purpose in creating a document determines whether the attorney-client privilege applies.

The court concluded that the bias-testing data was relevant as it had some probative value as evidence of disparate impact. This finding makes the privilege holding all the more consequential—the court shielded the data from disclosure even though it bore directly on the plaintiff’s claims, because counsel’s involvement in the testing process was sufficiently meaningful. The court found that “Workday has represented that its attorneys curated the data it used in the bias testing. The overall purpose of the testing was to provide legal advice and not to be used in a business capacity and it has not submitted the data to a regulatory body.” Without further explanation on how Workday’s counsel may have ‘curated’ the data, the court concluded that “Workday has shown more than mere direction by its attorneys” and found that the curated data was protected by the attorney-client privilege.

Our take

The court’s decision that the underlying data supporting the bias testing process is factual information not covered by the attorney-client privilege is consistent with well-established principles relating to attorney-client privilege. The court’s decision, however, also establishes key and relevant guidance that anti-bias testing created for the purpose of providing legal advice relating to NYC 144 can be shielded from disclosure by the attorney-client privilege. 

The important point here is that Workday’s lawyers claimed to be involved in determining which data was used in the testing and perhaps some of the testing methods. And, these materials were not disclosed to a regulatory body. Had the data been shared with a regulatory body, developed for another business purpose, analyzed by a non-legal external consultant or if Workday’s lawyers had not been meaningfully involved in the bias testing process, we expect the court would have reached a different conclusion.

While this decision relates specifically to Workday’s lawyers’ involvement in anti-bias testing in the context of Workday’s compliance with Local Law 144, the list of state regulations covering automated decision making in the employment context continues to expand. Even entities that are quite confident that they do not use AI to advertise job postings, screen applicants or assess employee performance, may be subject to enacted and pending regulations where they use technology to facilitate human decision making on employment matters, including requirements to engage in adequate validation of these technologies.

The takeaway from Mobley is straightforward: privilege can protect AI governance efforts, but the protection depends on counsel playing a substantive role in the testing process and not simply signing off on results after the fact. With that in mind, employers should consider the following steps to strengthen the privilege position of their AI compliance programs:

• Ensure that legal counsel directs the selection and curation of data used in bias testing, rather than simply reviewing results generated by technical teams or external vendors.
• Contemporaneous documentation matters. Employers should maintain records establishing that bias testing was undertaken to obtain legal advice, not for business or operational purposes.
• Sharing bias-testing results with external consultants, auditors or regulatory bodies risks waiving privilege. Workday itself did not claim privilege over testing conducted by an outside consultant.
• Underlying factual data will likely remain discoverable. As the Mobley court demonstrated by compelling production of Workday's federal demographic workforce reports, privilege does not extend to the raw data or to reports filed with government agencies, even where the analysis built on that data is protected.
• As the volume of AI-related legislation continues to surge across federal and state legislatures, entities should critically assess whether their existing AI governance frameworks are structured to preserve privilege, and ensure they are working with counsel well-versed in this rapidly evolving area.