Workday Discovery Dispute Is Pivotal Moment for AI Bias Lawsuits

Job seekers bringing a closely watched bias lawsuit targeting Workday’s AI-powered hiring software are fighting to uncover information needed to make or break their claims, and inspire similar cases.

Lead plaintiff Derek Mobley scored early-stage wins in the case alleging the software discriminated against him based on age, race, and disability. A California federal court declined to dismiss his disparate impact claims in 2024 and in May conditionally certified a collective action covering age bias.

AI-focused hiring bias disputes are in their infancy, and Mobley’s case has advanced much further than others, entering the evidence gathering phase known as discovery. The plaintiffs’ obtaining substantial discovery would boost the outlook for more cases, which attorneys view as difficult to prove given job applicants’ lack of insight into how AI evaluates them.

Mobley and opt-in plaintiffs are requesting—and Workday is largely opposing—access to demographics of workers who applied for jobs through the company’s platform, results of internal audits on its hiring tools, and technical details about how its algorithms operate.

Briefs on the discovery dispute are due ahead of an Aug. 21 hearing.

“It’s such a gigantic bite,” said Andrew L. Scroggins, an attorney at Seyfarth Shaw LLP. Going directly after Workday, rather than individual employers to whose jobs Mobley applied, poses unusually big challenges for plaintiffs.

Workday said in its briefs there could be millions of job applicants in Mobley’s proposed class, estimating its platform processed roughly a quarter of all job openings in a snapshot of US hiring from May 2023.

Judge Rita F. Lin of the US District Court for the Northern District of California ruled in July 2024 that Workday could be an agent of the employers using its tools, which subjects it to liability under federal anti-discrimination laws.

‘Always An Immediate No’

Mobley’s long-running job search began after his layoff from a tech company in 2017, he told Bloomberg Law in a May interview. He applied for more than 100 jobs through Workday’s platform mostly in financial, IT helpdesk, and customer service roles, while working temp positions.

He encountered persistent rejections, often coming from a Workday email address and sometimes within hours of submitting late-night applications, he said. This pattern led him to suspect the rejections were AI-generated.

“When it came to anything that was Workday-branded, it was always an immediate no,” said Mobley, who lives in Charlotte.

He said his decision to contact Birmingham, Ala., attorney Roderick T. Cooks in 2020 was partly inspired by watching the documentary “Persona: The Dark Truth Behind Personality Tests” about bias in job candidate evaluations.

Mobley alleges the way Workday’s platform evaluates applicants puts him at a disadvantage because he’s Black, over 40, and suffers from depression and anxiety. His performance on AI-scored cognitive tests and listing his 1995 graduation from the historically Black school Morehouse College might have been factors, he said.

Workday countered that the companies using its technology make hiring decisions, not the software itself.

“These tools look only at the skills listed in a candidate’s job application and compare them with the skills the employer has identified as needed for the job,” Workday spokesperson Connor Spielmaker said by email.

“They are not trained to use—or even identify—protected characteristics like race, age, or disability,” he added, noting the court dismissed Mobley’s intentional discrimination claims.

Customer, Employer Data

The plaintiffs contend the huge amount of data and job applicants involved shouldn’t prevent Workday from turning over what’s requested.

“That’s what a class action is for, to aggregate as many claims as you can so you can get a final answer to a central question,” Cooks said. “If a million people suffered the same harm, do you want a million cases to clog up the court system or do you want one adjudication?”

The applicant data is essential to determining whether some category of candidates faced a disadvantage, said Erica G. Wilson, an attorney at Fisher & Phillips LLP.

“For disparate impact analysis, you’re doing statistics,” she said.

If one demographic category of the candidate pool is selected less than 80% as often as the most selected category, then that’s evidence of disparate impact.

“But you can’t do that math if the data doesn’t exist,” Wilson said.

Workday argues the information is proprietary or company trade secrets, protected by attorney-client privilege, or customer property that Workday can’t access or divulge.

The privilege fight isn’t unique to AI bias cases, Wilson said. In traditional suits, privilege tends to cover emails to and from a company’s lawyers and final audit reports if they’re produced with attorney oversight.

“To me, the right answer is probably the report and communications are protected, but there’s nothing we can do to protect you with the raw data part,” she said.

The plaintiffs could subpoena applicant data from companies where Mobley applied or opt-in plaintiffs sought work, an option Workday suggested in its filings, Scroggins said—not an unusual maneuver, but one that would likely narrow class size and the sample for analyzing disparate impact.

“The larger the pool they’re looking at, the more likely it is they’ll find a statistically significant outcome,” he said.

Mobley’s filings hint he’s targeting at least a dozen companies, asking Workday to turn over applicant data from Comcast Corp., Duke Energy, Hewlett Packard Enterprise, and Unum Group, among others.

Even if plaintiffs get access to all applicant data connected to Workday’s platform, Workday could argue the disparate impact analysis should be done at the decision-making level by examining employer companies or divisions within them, Scroggins said.

Getting applicant data also likely won’t be enough for the plaintiffs to win, Wilson said. They’ll need evidence of how Workday’s tools are responsible for any bias that shows up in the data.

“There’s going to be two sides of this,” she said. “What is the impact that the Workday software itself is having, and then how much of that is beyond Workday’s control?”

The case is Mobley v. Workday, N.D. Cal., No. 3:23-cv-00770.