The AI Applicant Screening Software Made the Rejection. Now a Judge Says It Can Be Sued For It.

If you’re hiring in 2026, stop trusting the applicant screening tool and start reading actual resumes. Why? Yesterday, Judge Rita Lin of the U.S. District Court for the Northern District of California refused to dismiss most of the amended claims in Mobley v. Workday, a case that’s been working its way through federal court since February 2023. Workday sells AI-based applicant screening software that, according to the lawsuit, too often rejects older job seekers, racial minorities, and people with disabilities.

Derek Mobley, who first brought the case, knows the feeling of applying for a job he’s easily qualified for and getting an automated rejection in return. Court filings put the number at more than 100 rejections across companies that all used Workday’s hiring platform, with one rejection landing 55 minutes after he submitted the application, at 1:50 in the morning. Nobody was reading that file. Now there’s a collective action against Workday, certified under the Age Discrimination in Employment Act back in May 2025, arguing all of that adds up to discrimination. Yesterday’s ruling means the case keeps moving forward toward trial rather than getting tossed out.

And there’s a wrinkle in this one that a lot of people miss when they first hear about it: Mobley’s complaint isn’t against the employers who used Workday’s products at all. It’s against Workday itself, the software company, not a single business that ever actually hired or rejected him.

Workday tried to get the case thrown out, arguing the company doesn’t actually hire anyone, it just makes the software. It doesn’t sign offer letters. It doesn’t post the jobs. So how can it be liable? Lin wasn’t convinced. Because Workday builds and controls the algorithms that screen people out, she ruled the company can be treated as an “agent” of the employer under California’s Fair Employment and Housing Act, and separately allowed claims to proceed under the federal Americans with Disabilities Act.

This flips a basic assumption in employment law on its head. For decades, the rule was simple: if there’s a hiring problem, blame the employer. Now the software vendor is fair game too.

The lawsuit targets how these algorithms work under the hood. Nobody claims Workday’s code explicitly says “reject people over 40.” Instead, the system allegedly leans on proxy signals, things like gaps in a resume, graduation dates, or career paths that don’t look standard, and uses them to rank or filter people out before anyone reads the actual application. On paper, filtering for those things looks like a neutral business practice, just sorting data the way a spreadsheet sorts rows. In practice, the claim is that it systematically pushes older workers, women, and people with disabilities out before a human ever sees the application. In legal terms, that’s disparate impact, and intent doesn’t even need to come into it. A lopsided outcome on its own can be enough to get sued.

There’s a question worth asking Workday directly, and it’s one the lawsuit itself hasn’t fully answered yet: does the company actually know what its own algorithm is doing? Buried in the complaint is a claim about one of Workday’s tools, the Assessment Connector, that it picks up on patterns. If an employer tends to pass over candidates from a protected class, the system allegedly notices and starts recommending fewer of them, on its own, without anyone telling it to. That’s not a fixed checklist running the same way every time. That’s a system that learns from what it sees and adjusts. Workday’s defense has mostly leaned on the opposite idea, that it just runs the math an employer hands it and nothing more. Which version is true is still getting fought out in discovery, and frankly that might be the most uncomfortable part of this whole case: a company defending a system it may not be able to fully explain, even internally.

Here’s the part worth sitting with for a second. A human reading that same resume might look at a five-year gap and think layoff, caregiving, burnout recovery, any number of ordinary things, and decide to bring the person in anyway because something else on the page caught their eye. People notice the weird career pivot and get curious about it instead of penalizing it. An algorithm trained to flag deviations from a “normal” path doesn’t have that instinct, especially not one that’s quietly learning which deviations a given employer doesn’t like. It just sees a pattern that doesn’t match and moves on. The whole pitch behind automated screening was that it would cut bias out of hiring. There’s a real argument that it does the opposite, since a person can find a reason to rule someone in, and a model trained to spot anomalies, or worse, trained to mirror what an employer already wants, mostly just finds reasons to rule people out.

Workday says its tech only looks at qualifications and that the company tests for bias as part of a Responsible AI program. Sure. Every vendor in this space says some version of that. It doesn’t mean much without someone outside the company checking the work, and right now that someone outside checking the work is a federal court, not a regulator, not an internal audit team. That’s a strange place for an industry this big to end up.

Worth a side note here. New York City already tried to get ahead of this, sort of, with Local Law 144, which has required employers to run an independent bias audit on any AI hiring tool before using it, and post the results publicly, since 2023. Different approach entirely. Audit first, deploy second, and the obligation sits with the employer, not whoever built the tool. Mobley v. Workday goes after the thing LL 144 doesn’t touch at all, the vendor itself, after the fact, through a lawsuit instead of a filing requirement. A company could be fully compliant in New York and still get pulled into exactly this kind of suit somewhere else. Worth noting too that NYC’s own follow-through on LL 144 hasn’t been great, a state audit a while back found the city had barely caught any of the companies that weren’t actually complying. So “we follow Local Law 144” and “we’re covered” aren’t really the same sentence.

AI hiring software was sold on a promise: faster screening, more consistency, less human bias creeping into decisions. This case is testing whether that promise actually held up, or whether it just moved the bias somewhere harder to see, and dressed it up as efficiency along the way. The tech industry has spent years leaning on “black-box” algorithms as a shield, telling everyone the system is too complicated to second-guess. Courts are starting to ask for the source code anyway.