From Shubhi K | Product & Market Analysis
Mobley v. Workday: Where the AI Hiring Age Discrimination Case Stands Now
On this page
Workday told a federal court that about 1.1 billion applications were rejected through its software during the period at issue in Mobley v. Workday. That number is why an AI hiring lawsuit filed by one job seeker in February 2023 now covers a nationwide age discrimination collective. In October 2026 the case is still in discovery, with no ruling yet on whether any Workday tool actually discriminated against anyone.
Key takeaways
- A software vendor can be sued as the employer's agent. In July 2024 Judge Rita Lin held it plausible that employers delegated their traditional function of rejecting or advancing candidates to Workday's screening tools, which brings the vendor inside federal anti-discrimination law.
- The age claim became a nationwide collective in May 2025. The court conditionally certified an ADEA collective of applicants aged 40 and over, after Workday itself represented that about 1.1 billion applications had been rejected through its tools.
- The case has widened, not narrowed, in 2026. A June 2026 order kept California claims alive because Workday runs its tools from its California headquarters, and plaintiffs moved in September 2026 to certify subclasses by race, sex, age and disability.
- Public evidence of vendor contract changes is thin. The visible shift is in positioning and audit documentation, not in published contract terms, so buyers should not assume their own vendor terms moved.
What Mobley v. Workday is actually about
Derek Mobley applied for jobs at employers that used Workday's applicant platform and was rejected, by his account more than 100 times. He alleged that Workday's screening tools disproportionately rejected applicants who were Black, over 40, or disabled. The case number is 3:23-cv-00770 in the Northern District of California, before Judge Rita F. Lin.
The novelty is the defendant. Discrimination suits normally name the employer that made the hiring decision. Mobley sued the software vendor whose tools sat between him and those employers. He argued that the vendor itself was covered by Title VII, the Age Discrimination in Employment Act (ADEA) and the Americans with Disabilities Act (ADA).
Most of the claims are disparate impact claims. Disparate impact means a facially neutral practice produces worse outcomes for a protected group, whatever anyone intended. That matters because the plaintiffs do not need to show that Workday wanted to exclude older applicants, only that its tools did so and the practice cannot be justified.
My reading is that this is the most important pending case in candidate assessment AI, and it is important for a reason that has nothing to do with its eventual verdict. It has already settled, at the pleading stage, that a vendor can be a defendant. Every vendor contract and every procurement review now runs in the shadow of that ruling.
Direct answer: where does the case stand?
As of October 2026, Mobley v. Workday is in discovery and pre-certification. An ADEA collective of applicants aged 40 and over was conditionally certified in May 2025. Plaintiffs filed a broader class certification motion in September 2026, with a hearing reported for March 2027. No court has ruled on liability.
The Mobley v. Workday docket, February 2023 to October 2026
The docket is long, and the secondary coverage of it disagrees on several dates. The timeline below uses the dates given in law firm summaries of the orders themselves, and flags where a date rests on case trackers alone. Where sources conflicted, I used the one that cites an ECF number.
2023 and 2024: the pleadings fight
Mobley filed in February 2023. On 19 January 2024 Judge Lin granted Workday's first motion to dismiss with leave to amend, according to Proskauer's summary of the order. The court found the complaint too thin on which specific practice, and whose use of it, caused the alleged disparity.
Mobley amended. In April 2024 the Equal Employment Opportunity Commission (EEOC) filed an amicus brief arguing that Workday could be covered as an employment agency, an indirect employer or an agent of its customers. The EEOC took no position on the merits. On 12 July 2024 the court ruled on the amended complaint, and that order is the one every vendor lawyer now cites.
2025: the age collective and HiredScore
On 16 May 2025 Judge Lin conditionally certified the ADEA claim as a nationwide collective. A collective action differs from a class action: members must opt in rather than being included by default. The court found a plausible "unified policy: the use of Workday's AI recommendation system to score, sort, rank, or screen applicants".
Workday argued that HiredScore, a company it acquired in April 2024, was a separate product on separate technology. Coverage of a July 2025 order reports the court disagreed and treated HiredScore's Spotlight and Fetch features as part of Workday's application platform for collective purposes. In August 2025 the court let Workday give a list of employers that enabled those AI features to a third-party administrator, rather than disclose its general customer list.
2026: four plaintiffs, California law and a class motion
Plaintiffs' counsel reported that court-authorised notice went out in early 2026 with a 7 March 2026 response deadline. Three new named plaintiffs joined, bringing sex discrimination claims under Title VII and race, sex and age claims under California's Fair Employment and Housing Act (FEHA).
On 22 June 2026 the court largely denied Workday's motion to dismiss the third amended complaint. On 14 September 2026, according to the Warden AI tracker and Duane Morris, plaintiffs moved to certify four subclasses covering Black, female, older and disabled applicants. On 7 October 2026 a magistrate judge reportedly let Workday depose 15 of the 38 applicants who filed supporting declarations.
| Claim or theory | Status | Ruling |
|---|---|---|
| ADEA age disparate impact, applicants 40 and over | Alive, conditionally certified collective | 16 May 2025; applicant coverage reaffirmed in a March 2026 order. |
| Title VII and ADA disparate impact | Alive | 12 July 2024 order on the amended complaint. |
| FEHA claims (California) | Alive, on a California nexus theory | 22 June 2026, ECF No. 360. |
| ADA claim based on proxy indicators of health limits | Alive for one plaintiff | 22 June 2026. |
| Intentional discrimination | Dismissed, may return if discovery shows intent | 12 July 2024. |
| Workday as an "employment agency" | Rejected | 12 July 2024. |
| Race claim for an Asian American plaintiff; Workday as direct employer | Dismissed or stricken as outside the leave granted | 22 June 2026. |
Compiled from law firm summaries of each order. "Alive" means the claim survived a pleading challenge. It says nothing about whether the claim will win.
The agent theory the court accepted, in plain terms
Title VII, the ADEA and the ADA define "employer" to include any "agent" of an employer. That word is old. The July 2024 order is the first prominent ruling to apply it to an algorithmic screening vendor, and it is the reason this case matters beyond Workday.
The court held, as Cooley summarised it in July 2024, that a third-party agent "may be independently liable as an employer where the agent has been delegated functions traditionally exercised by an employer". Mobley plausibly alleged that employers had delegated their traditional function of rejecting candidates or advancing them to the interview stage.
Why delegation, not software, is the test
The court did not say that every HR software vendor is an employer. The theory turns on what the tool does. A tool that stores applications or schedules interviews performs a clerical task. A tool that decides who moves forward performs a hiring function the employer used to perform itself.
The court also pointed to Mobley's allegation that rejection emails arrived in the middle of the night, minutes after he applied. That detail supported an inference that no human reviewed the application. If your vendor's tool can reject a candidate with nobody looking, the delegation argument writes itself.
I think vendors have misread this part of the ruling. The common response is that "the customer makes the final decision." That is a true statement about the last step and an irrelevant one about the rejected majority, who never reached a final decision at all.
Why the employment agency theory failed
The EEOC's preferred theory was that Workday is an employment agency. The court rejected it. An employment agency procures workers for employers or job opportunities for workers, and the court found Mobley's allegation that Workday's systems "source candidates" conclusory.
The distinction has practical weight. An agency theory would have covered vendors that find candidates. The agent theory covers vendors that screen them. A sourcing tool that only surfaces profiles sits further from the line than a ranking tool that filters applicants out. The June 2026 order extended the logic under California law, holding Workday directly liable for its "own engagement in FEHA-regulated activities on the employer's behalf", independent of whether its customers are liable.
The vendor's location now matters too. Because Workday allegedly designs and operates its tools from California, out-of-state applicants can bring California claims against it. Any vendor headquartered in California should read that as a statement about its own exposure, a point developed further in the breakdown of California AI laws that bind software vendors.
Why 1.1 billion rejections changes the age discrimination stakes
The 1.1 billion figure came from Workday, not the plaintiffs. Workday raised it to argue that notice to a collective of that size would be unmanageable, potentially reaching hundreds of millions of people. The court rejected that argument and certified anyway.
The figure is not a count of people or of discriminatory acts. It counts rejected applications across the relevant period, so one person applying to 100 jobs appears up to 100 times. Notice coverage reported in 2026 describes the collective as applicants who were at least 40 and applied through the platform from 24 September 2020 onward.
What the number does establish is scale. The ADEA protects workers aged 40 and over. A screening practice that touches a pool this size does not need a large per-application effect to produce a very large aggregate one. That is the structural reason AI hiring lawsuits are more dangerous than individual hiring complaints: the same rule applies identically to every applicant.
It also explains why the case moves slowly. Discovery now includes requests for application records from opt-in plaintiffs, and disputes over applicant data held in customers' own Workday tenants. Each customer that enabled the AI features becomes a source of evidence, even though none of them is a defendant.
What HR tech vendors have changed since the Mobley v. Workday rulings
The brief for this post asked what vendors changed in their contracts and documentation. The honest answer is that the public record shows changes in documentation and positioning, and almost nothing verifiable about contract terms. Vendor contracts are private, and no vendor I could find has published a redline attributed to this case.
Documentation: the "assists, does not decide" framing
Workday's public position has been consistent. In 2023 a spokesperson told a local broadcaster that the lawsuit was without merit. In 2026 coverage, Workday said its recruiting tools assess job qualifications, do not make hiring decisions, and undergo responsible AI testing.
Workday's datasheet on responsible AI at HiredScore states that HiredScore's general manager supported New York City's bias audit rules in testimony, and that HiredScore has followed bias auditing practice for several years. That is a reasonable thing to publish. It is also, notably, the factual claim the agent theory puts in dispute. "Our tool does not decide" is a product description that a court will test against how customers actually configured it.
Bias audits are themselves becoming evidence. Case trackers report a July 2026 discovery order on Workday's internal bias-testing materials. If your vendor runs internal audits, assume those results are discoverable in litigation against the vendor. Local Law 144 audits and their limits are covered in the review of what NYC bias audits actually measure.
Contracts: what advisers say should change
Law firm commentary since the July 2024 order has pushed buyers in one direction: negotiate indemnification for algorithmic bias, and look hard at liability caps. A common enterprise software cap is a multiple of the prior 12 months of fees. That cap was set for outages and data loss, not for a share of a nationwide collective action.
I would not treat that advice as evidence that vendors have conceded anything. The incentive runs the other way. A vendor that accepts uncapped bias indemnity is pricing the risk the agent theory created, and few have a reason to do that before a liability ruling. The general mechanics of caps and carve-outs for AI vendors are in the analysis of AI agent liability caps and remedies.
| Term | Weak version | Version that holds up after Mobley |
|---|---|---|
| Decision rights | "Customer retains all hiring decisions" | Configuration states which stages can auto-reject, and customer can switch auto-reject off. |
| Bias testing | Vendor "tests for fairness" with no detail | Adverse impact results shared per role family, at a stated interval, with the method named. |
| Data for audits | Silent | Customer can export applicant-level scores and outcomes for its own audit. |
| Liability | Cap at 12 months of fees, no carve-out | Separate cap or carve-out for discrimination claims arising from the vendor's model. |
| Litigation cooperation | Silent | Defined process when a customer's applicant data is subpoenaed, as in Mobley. |
| Where the weak version wins | Low-volume hiring with human review at every stage. If nobody is auto-rejected, the delegation argument is weak and heavy terms add cost without reducing much risk. | |
Where this argument is weakest
Three parts of this analysis deserve less confidence than the rest.
First, nothing has been proven. Every ruling so far is about pleadings, certification or discovery. Conditional certification uses a lenient standard, and Workday can move to decertify after discovery, arguing that applicants rejected by different employers using different configurations are not similarly situated. That argument is strong. The court itself said material differences between scoring algorithms are best handled at decertification.
Second, the 18 standard deviation figure in the stat strip is a litigant's claim. Statistical disparity at that level would be striking, but the method, comparison pool and controls have not been tested by an opposing expert in public. I included it because it is what the plaintiffs will try to prove, not because it is established.
Third, the docket reporting is uneven. CourtListener and Clearinghouse refused automated access during research for this post, so the 2026 dates rest on law firm summaries and one case tracker whose own dates conflict with the firm summaries in places. Where they disagreed, I used the source citing an ECF number.
The case that the agent theory stays narrow
A reasonable reading is that the agent holding will be confined to tools that auto-reject without human review. Many vendors already sell ranking rather than rejection, and a recruiter who reviews a ranked list is making the decision. If courts draw the line there, the practical effect on most of the market is modest. I think that reading underestimates how often a ranked list is a rejection for everyone below the cut, but it is a respectable position.
What a buyer of candidate assessment AI should do now
This section is written for the person who owns recruiting operations. That is the head of talent acquisition or HR operations who signs the vendor renewal and runs the process. You do not need to predict the verdict. You need to know whether your configuration looks like delegation.
Start with the auto-reject question. Map every stage where a candidate can exit your funnel without a human looking at the application. Knockout questions, minimum score thresholds and automatic dispositions all count. This is the single fact the agent theory turns on, and most teams cannot answer it from memory.
Then get the data. Ask your vendor for applicant-level outcomes by stage so that you can run your own adverse impact check on age, the protected class in the certified collective. The four-fifths comparison is crude, but it is a start. A vendor that cannot export this data is asking you to carry risk you cannot measure, which connects directly to the questions in the AI agent procurement checklist.
Finally, check your funnel benchmarks against reality. If your screen-out rate at the first stage is far above peers, that is the place a disparity would show first. Comparable stage conversion data is collected in the ATS hiring process benchmarks.
My recommendation is blunt. If a stage auto-rejects and you have never measured its age impact, switch the auto-reject to a human review queue until you have. The cost is recruiter hours. The alternative is being the customer whose data defines the case.
Frequently asked questions
What is the current status of Mobley v. Workday?
As of October 2026, Mobley v. Workday is in discovery in the Northern District of California. An age discrimination collective was conditionally certified in May 2025. Plaintiffs moved in September 2026 to certify subclasses by race, sex, age and disability, with a hearing reported for March 2027. No court has decided whether Workday's tools actually discriminated against any applicant.
Can an AI hiring vendor be sued for discrimination?
Yes, on the reasoning in Mobley v. Workday. In July 2024 the court held it plausible that Workday acted as an agent of its customers because they delegated the traditional function of rejecting or advancing candidates to its tools. Federal anti-discrimination law defines employer to include agents, so a vendor performing that function can be a defendant. That is a pleading ruling, not a final judgment.
Who is eligible to join the Workday age discrimination lawsuit?
The conditionally certified collective covers applicants aged 40 and over who applied through Workday's platform and were denied employment recommendations. Notice coverage reported in 2026 says it reaches applications from 24 September 2020 onward. The opt-in deadline reported by plaintiffs' counsel was 7 March 2026. Anyone considering their position should check with the plaintiffs' counsel rather than rely on summaries.
What does the 1.1 billion figure in the Workday case mean?
Workday told the court that about 1.1 billion applications were rejected through its software tools during the relevant period. It raised the figure to argue the collective was too large to notify. It counts rejected applications, not unique people, and it is not a count of discriminatory decisions. The court certified the collective despite the argument.
Does Mobley v. Workday apply to HiredScore?
Coverage of a July 2025 order reports that the court treated HiredScore's Spotlight and Fetch features as part of Workday's application platform for the age collective. It did so despite Workday's argument that HiredScore was a separate product acquired in April 2024. The court left differences between scoring algorithms to be argued at the decertification stage after discovery.
What should employers using AI screening tools do after the Workday ruling?
Find every stage where a candidate can be rejected without human review, because that is what the agent theory turns on. Get applicant-level outcome data from your vendor and run an adverse impact check on age and other protected groups. Review whether your contract caps discrimination liability at a multiple of annual fees, and define how the vendor responds if your applicant data is subpoenaed.
Where to start this week
Book one hour with whoever configured your applicant tracking system. Ask a single question: at which stages can a candidate be rejected with no person reading the application? Write the answer down, with the setting that controls each one.
If the list is not empty, send your vendor one request in writing for stage-level outcome data split by age band. How quickly and completely they answer will tell you more about your exposure than any summary of this docket.
Related on regulation
State law is moving alongside the courts. See where the Colorado AI Act stands after SB 189 for the deployer obligations that apply to hiring tools.
References
- Proskauer, Law and the Workplace, Judge grants Workday's motion to dismiss in groundbreaking AI class action lawsuit, January 2024. Used for the 19 January 2024 order.
- Duane Morris Class Action Defense Blog, EEOC weighs in on novel artificial intelligence suit, 22 April 2024. Used for the EEOC amicus brief and its three coverage theories.
- Cooley, California federal court clears path for software developers' potential employment discrimination liability, 24 July 2024. Used for the agent holding and the rejected agency theory.
- Proskauer, AI bias lawsuit against Workday reaches next stage as court grants conditional certification of ADEA claim, June 2025. Used for the 16 May 2025 order and the 1.1 billion figure.
- HR Dive, Judge allows Workday to avoid disclosing full customer list, 28 August 2025. Used for the employer list ruling and the HiredScore features.
- Duane Morris Class Action Defense Blog, California federal court grants in part and denies in part Workday's motion to dismiss, 24 June 2026. Used for the 22 June 2026 order, ECF No. 360.
- Duane Morris, The Class Action Weekly Wire, Episode 166, 25 September 2026. Used for the class certification motion and the alleged disparities.
- Warden AI, Mobley v. Workday case tracker, accessed 28 October 2026. Used for the September and October 2026 docket events and the reported March 2027 hearing.
Weakest point in the source base: every source here is a law firm summary or tracker, not the court orders themselves, because the public docket mirrors refused automated access during research. Last verified 28 October 2026. Re-check the docket before relying on any 2026 date.
Related reading