1. What Is the Workday AI Lawsuit about?
Mobley v. Workday, Inc. .s pending in the U.S. District Court for the Northern District of California. The plaintiffs challenge how Workday's applicant-screening technology allegedly evaluates, ranks, and rejects candidates for employers using its recruiting platform.
The case reaches beyond one software provider. It tests how employment discrimination law applies when employers delegate parts of recruiting or applicant screening to an AI vendor.
Allegations against Workday'S Hiring Tools
The plaintiffs allege that Workday's automated screening tools caused applicants from protected groups to face discriminatory screening outcomes. Their allegations involve automated assessments, candidate ranking, repeated or rapid rejections, and criteria that allegedly function as proxies for protected characteristics.
The litigation includes claims involving race, sex, age, and disability under federal and California law.
These remain allegations. Workday disputes that its products discriminate and maintains that its recruiting technology evaluates job qualifications rather than protected characteristics.
Orders allowing claims to proceed do not establish that unlawful discrimination occurred. They determine whether the claims may continue to later stages of litigation.
Why AI Vendor Liability Is a Central Issue
A central question is whether an AI vendor can face liability when employers delegate parts of the hiring process to its technology.
The court rejected the theory that Workday qualified as an employment agency under the federal statutes at issue, but it allowed certain federal disparate-impact claims to proceed under an agency theory. At the pleading stage, allegations that employers delegated traditional hiring functions such as applicant screening were sufficient to keep that theory in the case.
A June 2026 ruling also allowed California FEHA claims to continue. The court found the alleged California connection sufficient at the pleading stage while leaving the merits of those claims unresolved.
These rulings do not make every software vendor an employer or statutory agent. The analysis can depend on the function the vendor performs, how much authority an employer delegates, how the system is configured, and whether the technology merely provides information or meaningfully participates in screening decisions.
ADEA Collective Action Vs. Rule 23 Class Certification
The ADEA collective and the pending Rule 23 motion are different procedures.
In 2025, the court preliminarily certified an ADEA collective under the Fair Labor Standards Act's collective-action mechanism for certain age-discrimination claims.
In 2026, the plaintiffs separately sought certification of proposed subclasses involving race, sex, age, and disability claims. Those Rule 23 issues remain pending, with a hearing scheduled for March 2027.
Neither preliminary collective certification nor Rule 23 certification establishes that discrimination occurred. Certification addresses whether claims may proceed on a group basis, not whether the plaintiffs ultimately prove liability.
2. When Can AI Hiring Bias Create Employment Discrimination Liability?
AI employment discrimination claims can arise under different legal theories depending on the protected characteristic and the screening practice being challenged.
Using artificial intelligence is not itself unlawful. The analysis turns on the specific employment practice, the resulting disparity, causation, and any defense or justification available under the governing statute.
Disparate Treatment and Disparate Impact
Disparate treatment generally concerns intentional discrimination based on a protected characteristic.
Disparate impact addresses a facially neutral employment practice that disproportionately harms a protected group without satisfying the legal standards applicable to that practice.
In an automated hiring case, the useful sequence is:
Specific employment practice → measurable disparity → causation → statutory defense
The challenged practice could be a knockout rule, assessment score, ranking threshold, résumé-screening criterion, or another identifiable part of the hiring process.
| Law | Core AI Hiring Question |
|---|---|
| Title VII | Did a specific screening practice cause the disparity, and is it job related for the position and consistent with business necessity? |
| ADEA | Did the challenged practice adversely affect applicants age 40 or older, and can the defendant establish an applicable reasonable-factor-other-than-age defense? |
| ADA | Did a qualification standard screen out a qualified individual with a disability, and was a reasonable accommodation available where required? |
| California FEHA | Did an automated decision system or proxy cause discrimination based on a protected characteristic, and what evidence exists regarding testing, job relevance, and preventive measures? |
Title VII
- Core AI Hiring QuestionDid a specific screening practice cause the disparity, and is it job related for the position and consistent with business necessity?
ADEA
- Core AI Hiring QuestionDid the challenged practice adversely affect applicants age 40 or older, and can the defendant establish an applicable reasonable-factor-other-than-age defense?
ADA
- Core AI Hiring QuestionDid a qualification standard screen out a qualified individual with a disability, and was a reasonable accommodation available where required?
California FEHA
- Core AI Hiring QuestionDid an automated decision system or proxy cause discrimination based on a protected characteristic, and what evidence exists regarding testing, job relevance, and preventive measures?
Under Title VII, a practice that satisfies the business-necessity standard may still face scrutiny if a less discriminatory alternative would serve the employer's legitimate needs and the employer refuses to adopt it.
ADEA Applicant Claims Require Additional Context
In Mobley, the Northern District of California has held that job applicants may pursue disparate-impact claims under ADEA § 623(a)(2).
Workday sought interlocutory review of that interpretation, but the court denied certification under 28 U.S.C. § 1292(b) in July 2026.
That position should not be stated as though every court has treated applicant coverage under § 623(a)(2) identically. For an employer facing an age-based AI screening claim, forum and controlling precedent may affect the analysis.
Disability Screening and Accommodation Risks
Disability issues can arise before a final hiring decision.
An automated assessment may evaluate speech patterns, reaction time, cognitive performance, physical behavior, work-history patterns, or other characteristics that affect candidates with certain disabilities differently. An application system may also screen out someone who could perform the job with a reasonable accommodation.
The Workday plaintiffs have alleged, among other theories, that certain indicators could operate as disability proxies. Whether a variable actually functions that way requires evidence.
Employers also need a workable process for applicants who request an accommodation during an assessment or automated screening step.
3. How California Regulates AI Hiring under FEHA
California's employment regulations expressly address automated decision systems used in recruitment, screening, hiring, promotion, and other employment decisions.
The regulations took effect on October 1, 2025. They do not prohibit AI in employment. Instead, they clarify how existing Fair Employment and Housing Act protections apply when an automated system participates in an employment decision.
Automated Decision Systems and Protected Characteristics
California defines an automated decision system broadly as a computational process that makes or facilitates a decision concerning an employment benefit. Artificial intelligence, machine learning, algorithms, statistical processes, and related data-processing techniques can fall within that definition.
An employer or other covered entity may not use an automated decision system or selection criterion in a manner that violates FEHA.
The analysis can include criteria that appear neutral but correlate closely with a protected characteristic.
California also treats anti-bias testing and related preventive efforts as potentially relevant evidence. The existence or absence of testing, its quality, scope, timing, results, and the response to those results may all be considered.
This does not mean California requires every employer to conduct a bias audit, nor does completing an audit create an automatic affirmative defense.
Agents, Proxies, Bias Testing, and Records
An agent may include a third party performing functions traditionally exercised by an employer, including applicant recruitment, screening, hiring, or other employment decisions.
Whether a vendor fits that definition depends on the actual relationship and conduct. A contract label alone does not resolve the issue.
Proxy criteria also require scrutiny. A variable that does not expressly identify race, disability, age, sex, or another protected category may still raise a discrimination issue if it is closely correlated with a protected characteristic and affects an employment decision.
California generally requires covered employers to preserve relevant employment records, including certain automated-decision-system data and selection criteria, for four years from the date the record was made or the date of the personnel action involved, whichever is later.
That retention rule can extend the preservation period beyond four years from the date a particular record was first created.
4. What Evidence Matters in an AI Hiring Discrimination Case?
Applicant outcomes may reveal a disparity, while technical records can help identify where the challenged decision occurred.
The strongest analysis connects a specific screening practice to the applicants affected by it rather than treating the entire hiring platform as one undifferentiated system.
Applicant Outcomes and Selection Data
Relevant records can include:
Applicant pools for comparable positions;
Candidates who advanced or were rejected;
Selection and rejection rates;
The stage at which applicants left the process;
Job-specific qualifications;
Assessment scores or reason codes;
Accommodation requests;
Screening criteria used during the relevant period.
The comparison group must be meaningful. Combining unrelated positions, different qualification standards, or materially different hiring periods can obscure the effect of the challenged practice.
A statistical disparity also does not establish causation by itself. The evidence must connect that disparity to the employment practice being challenged.
Vendor Data, Bias Testing, Privilege, and Human Review
Reconstructing an automated rejection may require model versions, knockout rules, customer settings, ranking criteria, audit logs, human overrides, and historical configuration records.
The Workday litigation also shows why data possession and discovery control are different questions.
In a May 2026 discovery ruling, the court addressed attorney-client privilege over certain attorney-directed bias-testing work and whether Workday had Rule 34 control over customer applicant data. The court also required Workday to produce certain EEO-1 and OFCCP materials relevant to its knowledge of potential demographic disparities.
That ruling does not mean all bias testing is privileged or that vendors never control customer data. Privilege depends on why and how the work was created. Discovery control depends on the legal and practical ability to obtain the information.
A human reviewer appearing somewhere in the workflow does not by itself establish that a person made the challenged decision. The relevant question is where the applicant was screened out and which criterion produced that result.
5. What Should Employers Review after the Workday Litigation?
Employers using third-party AI tools should be able to reconstruct the decision chain from application through screening and final disposition.
The practical questions are what the tool does, which criteria it uses, who controls those criteria, what information the employer receives, and what happens when an applicant is rejected.
AI Vendor Due Diligence and Contracts
Vendor review should go beyond a statement that a product is "bias free" or compliant with employment law.
Depending on the technology, employers may need to examine:
Which employment decisions the system influences;
Screening variables and knockout criteria;
Validation and bias-testing information;
Customer configuration options;
Accommodation procedures;
Access to applicant and outcome data;
Audit and logging capabilities;
Notice of material model changes;
Record-retention responsibilities;
Cooperation after an EEOC or CRD charge, discovery request, or lawsuit.
Contracts can allocate responsibilities between an employer and vendor, but contractual language does not necessarily eliminate statutory liability to applicants.
Hiring Workflow, Documentation, and Human Oversight
HR teams should know where automated screening occurs, what causes an applicant to advance or stop, which criteria apply to the position, and how accommodation requests are handled.
A model update can make an earlier rejection difficult to reconstruct unless the prior configuration was preserved. Employers therefore need records that show which model, scoring rule, or threshold applied when a disputed decision occurred.
Testing can identify unexpected outcome differences, but its value depends on what was tested, whether the test matched the deployed system, when it occurred, what the results showed, and what changed afterward.
Attorney review may become appropriate when an applicant or agency challenges a particular screening criterion, an EEOC or CRD charge arrives, or selection-rate data shows an unexplained disparity. A discrimination complaint that triggers preservation duties or a system whose historical configuration cannot be reconstructed can also require prompt legal review.
6. Frequently Asked Questions
No. The litigation remains pending, and the court has not determined that Workday's hiring tools unlawfully discriminated against applicants.
Several claims have survived dismissal. The case also includes a preliminarily certified ADEA collective for certain age claims, while separate Rule 23 class-certification issues remain pending. None of those procedural rulings is a finding that discrimination occurred.
Potentially.
An employer does not receive an automatic defense merely because a vendor supplied the screening technology. Liability depends on the applicable statute, the challenged practice, the employer's role in selecting or configuring the tool, the vendor relationship, and the evidence connecting the practice to the alleged discriminatory outcome.
A vendor may also face separate liability if the governing law and facts support it.
The plaintiffs sought Rule 23 certification in September 2026 for proposed subclasses involving race, sex, age, and disability claims. A hearing on class certification is scheduled for March 9, 2027.
The court's certification decision will address whether those claims may proceed on a class basis. It will not determine whether Workday ultimately violated employment discrimination law.
No. Bias testing is not an automatic defense.
California treats the existence or absence of anti-bias testing, together with factors such as its quality, recency, scope, results, and the employer's response, as potentially relevant evidence.
A test performed on a different model, outdated configuration, or materially different hiring process may have limited value in evaluating the system actually challenged.
05 Oct, 2026
