> This is the markdown version of https://www.classet.ai/blog/workday-ai-hiring-lawsuit > Learn more at https://www.classet.ai ![Low-poly gavel resting over a stream of job applications flowing into an algorithmic black box](/_next/image?url=%2Fimages%2Fblog%2Fworkday-ai-hiring-lawsuit.png&w=3840&q=75) # What the Workday AI Lawsuit Means for Your Hiring A federal court is letting an AI hiring vendor be sued for discrimination. Here's what Mobley v. Workday changes for employers, and how to screen without the risk. [![Paul Jones](/_next/image?url=https%3A%2F%2Fassets.basehub.com%2Fe0b5701f%2F6599306507912123f90f150a8bfaaf6c%2Fscreenshot-2026-01-28-at-10.53.16-am.png%3Fwidth%3D100%26height%3D100%26quality%3D100&w=96&q=75) Paul JonesHead of Growth at Classet ](/blog/authors/paul-jones) June 12, 2026 AI Recruiting, Hiring Tips A federal judge in California decided something most of the HR software industry assumed could never happen: the company that built the AI screening tool, not just the employer who bought it, can be sued for discrimination. The case is _Mobley v. Workday_, and it has been grinding through the Northern District of California since 2023. If you run hiring through any tool that scores, ranks, or auto-rejects applicants, the question it raises is no longer academic. It's whether the way your stack makes decisions would survive the same scrutiny. Quick answer In _Mobley v. Workday_, a court ruled the AI hiring vendor can be held liable as an "agent" of the employers using it, because its tools were "participating in the decision-making process" rather than applying employer criteria "in a rote way." In May 2025 the court conditionally certified a nationwide collective of applicants 40 and older screened by Workday's AI since September 24, 2020. The risk isn't using AI to screen. It's letting a model reject people on its own, with no human reading the result and no record of why. ## What the lawsuit actually alleges Derek Mobley, a Black man over 40 with diagnosed anxiety and depression, says he applied to more than 100 jobs at companies running Workday and was rejected from every one, sometimes within an hour of applying. He sued Workday itself, arguing its algorithmic screening tools sorted candidates in a way that produced disparate impact on race, age, and disability. The claim that matters is the legal theory, not the individual story. Mobley argues Workday isn't a neutral piece of software sitting between him and an employer. By running the advance-or-reject decision across hundreds of customers, Workday became an active participant in hiring, which means the anti-discrimination statutes apply to the vendor directly. 2023 Year the case was filed in N.D. Cal. July 2024 Court let the discrimination claims proceed 40+ Age of the nationwide collective certified in May 2025 Sept 24, 2020 How far back the collective reaches ## The line the judge actually drew This is the part worth reading twice, because it tells you exactly what to check in your own tools. When Workday moved to dismiss, the court refused. In her July 12, 2024 order, Judge Rita Lin wrote that Workday's software was "not simply implementing in a rote way the criteria that employers set forth, but is instead participating in the decision-making process by recommending some candidates to move forward and rejecting others." That sentence is the whole case. The court drew a line between two kinds of tools. On one side, a system that applies the explicit criteria an employer set, the way a spreadsheet filter would. On the other, a model that recommends, ranks, and rejects on its own logic. The first looks like a passive instrument. The second looks like an agent making employment decisions, and agents of employers are covered by Title VII, the ADEA, and the ADA. Then it scaled. In May 2025 the court conditionally certified a nationwide collective of every applicant 40 or older who was rejected through Workday's AI tools since September 2020. One alleged flaw in one vendor's model stopped being a single dispute and became a class action spanning hundreds of employers, because the same model touched all of them. ## The risk pattern, not the technology It's tempting to read this as "AI hiring is dangerous, slow down." That's the wrong takeaway, and it will cost you good candidates. The danger isn't that a tool uses AI. It's a specific pattern that happens to be common in older screening products: - A model assigns a score or rank using logic the employer can't fully explain. - It's trained on historical hiring data, which carries forward whoever got hired before. - It rejects candidates on its own, with no human reading the decision, sometimes in minutes. - Nobody, not the candidate and often not the employer, can say why a given person was screened out. Stack those four together and you get disparate impact that nobody chose and nobody can defend. Plaintiffs don't have to prove intent. They only have to show the numbers came out unevenly across a protected group, and then the burden shifts to you to prove the practice was job-related and necessary. That's hard to do when you can't explain how the tool decided. We saw the same failure mode in [Stanford's analysis of 4 million AI-screened applications](https://www.classet.ai/blog/ai-hiring-bias-stanford-study): the bias didn't come from a checkbox, it came from a model owning the reject decision at scale. The regulators are circling the same pattern. [NYC's Local Law 144](https://www.deloitte.com/us/en/services/audit-assurance/articles/nyc-local-law-144-algorithmic-bias.html) requires an independent annual bias audit and candidate notice for automated employment decision tools. Illinois, Colorado, and the EEOC are all converging on the same three demands: tell candidates, audit for impact, and keep a human in the loop on adverse decisions. ## How we built Classet so this can't happen We didn't design [Joy, our AI interviewer](https://www.classet.ai/), to make the call. We built her to do the part humans are bad at, which is talking to applicants at scale and engaging them when they're warmest, right after they apply, and then hand a clear summary to the person who decides. The difference maps almost exactly onto the line the Workday judge drew. A human is always in the loop. Joy phone-screens candidates, asks the structured questions you set, and hands your recruiters the transcript summary and the recording of every call. She does not autonomously reject anyone on a hidden score. A recruiter reads the summary, listens to the call when they want to, and makes the advance-or-reject decision with far more to go on than a resume. That equips your team to make their own calls faster and with better information, which cuts time to hire, and it keeps the decision where the law expects it to sit, with a person who can explain it. We screen against your criteria, not a personality model. Joy evaluates objective, job-related requirements and knockout questions you define, like a CDL, a certification, or availability for a shift. She doesn't rank people on tone, vibe, or inferred personality, which is precisely the kind of opaque scoring that turns into a proxy for age, race, or disability. The decision is legible. Because Joy produces a structured summary tied to the questions you asked, there's a record of why a candidate moved forward. That's the opposite of a black-box score, and it's what you'd want in hand if anyone ever asked you to defend a hiring outcome. We audit for impact, and we meet the standards regulators have already written. Classet is designed to meet or exceed AEDT standards including NYC Local Law 144, and we run independent third-party bias audits that monitor for disparate impact in aggregate. We use accepted proxy methods like BISG and geocoding for that aggregate auditing only, never to decide anything about an individual, which is the practice regulators recognize as valid for monitoring. None of that slows hiring down. It's faster to talk to applicants at scale and engage them when they're warmest, right after they apply, then let a recruiter act on a clean summary and recording than to wait on a resume pile. You just don't take on the one risk the Workday case is built around: a machine quietly saying no, at scale, with no human and no paper trail. ## Common questions about AI hiring and the Workday case Does Mobley v. Workday make AI hiring illegal? No. The case doesn't ban AI screening. It holds that a vendor whose tool participates in the hiring decision can be sued under the same anti-discrimination laws as an employer. Using AI to talk to candidates and summarize them is very different from using AI to autonomously reject them. The legal exposure tracks who, or what, owns the reject decision and whether the outcome can be explained. If I use an AI hiring tool, can I be sued too? The employer has always been on the hook for discriminatory outcomes from a selection tool, even one a vendor built, per longstanding EEOC guidance. What Mobley adds is that the vendor can be liable as well. So both ends of the chain are exposed. The practical defense is the same one the court hinted at: keep a human making adverse decisions, screen on job-related criteria, and be able to show why each candidate was advanced or rejected. What makes Classet different from the tools in the lawsuit? The tools at issue allegedly recommended and rejected candidates on their own. Joy doesn't. She conducts the screen against the objective criteria you set and hands a structured summary to a recruiter, who makes the call. Classet also runs independent third-party bias audits and is built to meet AEDT standards like NYC Local Law 144. A human owns the decision, the criteria are job-related, and the reasoning is on the record. How do I check whether my current screening tool is risky? Ask three questions. Does a model reject candidates on its own, or does a person? Is it scoring objective job requirements, or inferring traits like personality and tone? If someone challenged a rejection, could you explain why that candidate was screened out? If the honest answers are "the model," "traits," and "no," you're carrying the exact pattern this litigation is built around. ## Key points - _Mobley v. Workday_ established that an AI hiring vendor can be sued for discrimination as an "agent" of the employers using it, not just the employer alone. - The deciding factor was that the tool was "participating in the decision-making process" by recommending and rejecting candidates, rather than applying employer criteria "in a rote way." - In May 2025 the court certified a nationwide collective of applicants 40 and older screened by Workday's AI since September 2020, turning one model's flaw into a class action across hundreds of employers. - The real risk is a four-part pattern: opaque scoring, trained on historical bias, auto-rejecting with no human, and no record of why. - Classet avoids that pattern by design. Joy screens against your objective criteria and hands a structured summary to a human who makes the decision, backed by independent bias audits and AEDT compliance. ## See how it works If you're not sure whether your screening stack owns the reject decision, that's worth knowing before a candidate or a regulator asks. [Book a demo](https://www.classet.ai/demo) and we'll show you how Joy engages applicants at scale the moment they apply while keeping the decision, and the paper trail, with your team. [![Paul Jones](/_next/image?url=https%3A%2F%2Fassets.basehub.com%2Fe0b5701f%2F6599306507912123f90f150a8bfaaf6c%2Fscreenshot-2026-01-28-at-10.53.16-am.png%3Fwidth%3D100%26height%3D100%26quality%3D100&w=128&q=75) Paul Jones Head of Growth at Classet Paul comes from an operator background running an Alpine-owned company, and brings firsthand experience with the hiring challenges Classet was built to solve. He's driven by a belief that the right technology can make meaningful work more accessible. ](/blog/authors/paul-jones) ## Explore More ### Use Cases - [RPO / BPO Recruiting](/use-cases/call-centers-bpo) - [Healthcare Recruiting](/use-cases/healthcare) - [Hospitality Recruiting](/use-cases/hospitality) ### Integrations - [Greenhouse](/integrations/greenhouse) - [Bullhorn](/integrations/bullhorn) - [Lever](/integrations/lever)