One hundred rejections, some in the middle of the night
The story that is redefining the legal liability of artificial intelligence in employment begins with an ordinary job seeker. Derek Mobley, a professional with a Morehouse College degree and experience in finance, IT and customer service, submitted more than one hundred job applications through Workday’s platform; he was rejected every time —often within minutes, sometimes in the middle of the night— without a single interview. For Mobley, an African-American man with a disability and over 40, the pattern of instant rejections did not look like the work of human recruiters reading resumes, but of an automatic filter discarding him before anyone saw him.
That intuition became a lawsuit in February 2023, and its central argument is response time. Mobley’s retort is that he was sometimes rejected within hours or minutes of applying, suggesting it was Workday’s automated screening tools that were making the rejection decision on behalf of the employers. A human does not read, evaluate and discard an application at three in the morning in a matter of minutes; an algorithm does. On that inference the case is built: if the machine decides, do anti-discrimination laws reach the machine and the company that makes it?
The lawsuit does not allege that anyone at Workday wanted to discriminate, and that distinction is the case’s technical key. The claim centers on disparate impact: plaintiffs argue that Workday’s algorithm functionally deprioritized older applicants, effectively excluding a protected class from employment consideration; according to the complaint, the discrimination flowed from the software’s design, which ranked, screened and filtered in ways that penalized age with no human intervention. The disparate-impact doctrine does not require proving intent: it is enough to show that an apparently neutral practice disproportionately harms a protected group. Applied to an algorithm, the question stops being what the programmer intended and becomes what the program does.
The ‘agent’ theory: why the vendor is in the dock
The case’s initial legal obstacle was obvious: Workday was not Mobley’s employer. U.S. anti-discrimination laws —Title VII, the age-discrimination law (ADEA), the disability law (ADA)— are aimed at employers, and Workday only sells the software they use. The court’s answer to that obstacle is the case’s most consequential piece. The court rejected the argument that Workday was an ‘employment agency,’ but accepted the argument that Workday can be considered an ‘agent,’ because it was plausible that Workday’s customers had delegated to it their traditional functions of rejecting and advancing candidates.
The agent theory has a logic that is hard to dodge: if an employer delegates to software the function of deciding who advances and who is left out, the maker of that software assumes the delegated function, and with it the liability. The case’s procedural trajectory confirmed the theory holds. In July 2024, the court denied Workday’s second motion to dismiss, allowing the claims to proceed. And in May 2025 came the leap in scale: Judge Rita Lin granted conditional certification for Mobley v. Workday to proceed as a nationwide collective action under the age-discrimination law, meaning potentially millions of applicants aged 40 and over who were screened through Workday’s AI-powered tools since September 2020 can join the lawsuit.
The case entered its most active phase in 2026, with two milestones that widened its reach. A federal court authorized notice to potential class members in February, with a deadline of March 7, 2026 to opt in; the court authorized notice to individuals who applied for jobs through Workday’s platform since September 24, 2020 and were age 40 or older at the time of application. And weeks later, the defense’s last major argument fell. The case moved from conditional collective certification into active litigation of its legal scope, with a landmark March 6, 2026 ruling rejecting Workday’s strongest remaining dismissal argument: that the age-discrimination law does not cover job applicants. Plaintiffs responded on March 30 with an amended complaint reasserting the California state and disability claims.
The figure that sizes the case: 87 percent
If this litigation matters beyond its parties, it is because of the scale of the practice it examines. Algorithmic screening is not a technological rarity: it is the labor market’s standard. AI is embedded across the hiring landscape: 87 percent of employers now rely on it to evaluate candidates, but statutory frameworks have not kept pace with the scale or structure of automated decision-making. For the vast majority of today’s job seekers, the first evaluation of their candidacy —and frequently the decisive one— is made by an automated system.
The origin of the risk is documented by academic research, and the lawsuit incorporates it as a foundation. Researchers have recognized that AI is biased, developing implicit and explicit biases based on the training data it is provided; now, applicants like Mobley allege that AI relies on those biases when screening out candidates. A system trained on a company’s hiring histories learns that company’s patterns, including its biases: if it historically hired fewer older people, the model can learn that age correlates with rejection and replicate it at industrial scale. In seeking certification, Mobley claimed that Workday’s tools were ‘designed in a manner that reflects employer biases and relies on biased training data.’
The court validated, for certification purposes, that this common design constitutes a unified policy that can be challenged as a block. The court agreed this adequately alleged ‘the existence of a unified policy: the use of Workday’s AI recommendation system to score, sort, rank, or screen applicants,’ and found immaterial for certification purposes that class members’ qualifications varied, because the common injury was simply being ‘denied the right to compete on equal footing with other candidates.’ That formulation —the right to compete on equal footing— is what turns a software dispute into a civil-rights case.
The two readings of the case, each at its strongest
Like every lawsuit that may set precedent, this one admits two legitimate readings that deserve balanced presentation, starting with the defense’s. Workday denies the allegations, and its position has serious foundations: the company maintains that its tools do not make hiring decisions —employers configure the criteria and keep the final say— and that holding the vendor liable for the outcomes of configurations its clients control would discourage the development of technology that, used well, can reduce human bias rather than amplify it. It is worth recalling that the lawsuit’s allegations have not been proven at trial: what the courts have resolved so far is that the case may proceed, not that the discrimination occurred.
The reading of the plaintiffs, and of much of labor-law scholarship, is that the mass delegation of screening to algorithms created a liability vacuum this case has come to close. Mobley v. Workday may be the first suit of its kind, but others against AI hiring practices are already progressing through the courts: a group of job applicants is suing Eightfold AI with an even more novel claim, that AI employment tools should be subject to the Fair Credit Reporting Act. For this current, if no one answers for the filter —neither the employer that says it did not decide, nor the vendor that says it only sells software— anti-discrimination protections become empty precisely at the gateway to employment.
The case’s practical effect is already felt in the market, whoever wins. Once the court establishes that AI vendors can be held liable for discriminatory outcomes, every screening tool provider is on notice, and so is every company using them. Labor firms report a wave of audits of hiring systems, and the question coming from legal departments is always the same: can we show our filter does not penalize a protected group? The court also ordered Workday to turn over the list of its employer clients who used its tools, a procedural step that extends the company’s scrutiny to its entire market.
Who is in the lawsuit and what comes next on the calendar
Behind the legal theory there is a group of people with parallel stories, and their accumulation is part of the case’s force. Four additional plaintiffs, all over 40, have joined the case, describing similar patterns of near-instant rejections and alleging that Workday’s screening tools disproportionately excluded them based on age. Five different professional trajectories, one same pattern of automatic discard: that repetition is what transforms an individual grievance into the allegation of a systemic practice, and what persuaded the court that the group shares a common injury that can be litigated as a block.
The procedural window to join the collective action opened and closed this year, with mechanics that give a sense of the scale. The notice authorized in February invited anyone who had applied through the platform since September 24, 2020, at age 40 or older, to participate via a consent form with a March 7, 2026 deadline. How many people joined within that window is one of the figures that will define the litigation’s economic magnitude: a class of millions of members turns any damages calculation, or any settlement negotiation, into a number of another category.
With the class notified and the dismissal arguments exhausted, the case enters the terrain where these disputes are decided: discovery. The order requiring Workday to hand over the list of its employer clients points in that direction, because it lets plaintiffs reconstruct where and how the challenged tools operated. For the company, each discovery phase is also a reputational risk: internal documents on the design and testing of its algorithms can end up in the public record. It is the kind of pressure that, historically, pushes parties toward a settlement before trial.
What employment lawyers advise in the meantime
Without waiting for the outcome, the firms advising employers have already translated the case into a list of duties. The central recommendation is the audit: examine automated screening systems for disparate outcomes by age, race, gender or disability, before a plaintiff’s lawyer does. The difference between companies that survive legal challenges and those that don’t lies in systems that catch mistakes quickly and take corrective action. The defensive logic is simple: an employer that can show it monitors and corrects its filters has a radically better legal position than one that delegated blindly.
The other recurring piece of advice inverts these tools’ sales pitch. Yes, AI can screen thousands of resumes in seconds, and yes, it saves time and money; but if it’s creating discriminatory outcomes —even unintentionally— the lawsuit costs will dwarf any efficiency gains; the regulatory environment is clear: hiring is high-risk, and that means slower is sometimes better if it’s more defensible. That U.S. employment counsel recommends slowing down automation in the name of defensibility is, in itself, a data point on how this case already governs the market without need of a verdict.
The regulatory context: a moving mosaic
The litigation advances in parallel with intense and fragmented legislative activity on AI in employment. The United States has no federal algorithmic-hiring law, and the vacuum is being filled by the states, each in its own way: bias-audit requirements, obligations to notify candidates that an automated system will evaluate them, rights to human review. The result is a mosaic in which the same tool can be legal in one state and require safeguards in the neighboring one, multiplying compliance costs for employers hiring at national scale.
That fragmentation turns the Mobley case into something more than a lawsuit: the functional substitute for the regulation that does not exist. As long as Congress does not legislate, the frontier of what is permitted in algorithmic hiring is being drawn by the courts, case by case. And the direction of the line, so far, is consistent: liability is not diluted by automating. For the millions of people evaluated every year by systems that will never see them, that direction matters more than any vendor’s promise about the neutrality of its algorithms.
For Latin America, where the same screening platforms expand without an equivalent framework of litigation or audit, the case offers a preview of the debate to come. The tools that filter candidates in Bogotá, Mexico City or Panama City are frequently the same ones being examined in the California courtroom, trained on similar logics. The question of who answers when the machine filters has no answer yet in most of the region’s legislation, and the U.S. experience —including the cost of resolving it lawsuit by lawsuit— is study material for Latin American regulators who would rather get ahead of it.
The balance of the data
The Mobley case against Workday condenses the collision between two realities: that of a labor market where 87 percent of employers already use AI to evaluate candidates, and that of a legal framework written for a world where hiring decisions were made by people. The docket’s data —more than one hundred rejections, some within minutes and in the middle of the night; a potential class of millions of applicants over 40 screened since 2020; a March 6, 2026 ruling that kept the lawsuit alive by confirming the law also protects applicants— have taken the question of algorithmic liability further than any bill.
The verdict the data leave, with the trial still pending, is twofold. Legally, the agent theory has survived every dismissal attempt: the court accepts that whoever receives by delegation the function of deciding can answer for how it decides, whether human or software. Practically, the case has already changed market behavior before any judgment: vendors and employers audit their filters because they know the next lawsuit may carry their name. The essential remains to be resolved —whether the alleged discrimination occurred and who will pay for it— but the structural lesson is already written: automating a decision does not exempt it from the law that governed it when it was human. And that lesson, born in a California courtroom, travels with every hiring-software license sold in the rest of the world.