Rejected by an Algorithm at 1:30 AM: Inside Automated Hiring

A man named Derek Mobley applied for more than a hundred jobs through the same widely used hiring platform. He was rejected from every one. In at least one case, the rejection email arrived at 1:30 in the morning on a weekend, hours after he'd applied, far too fast for any human to have read the application.

Mobley, who is Black, over 40, and has a disability, came to believe the pattern wasn't bad luck. He believed the AI screening tool was filtering him out before a human ever saw his name. So he sued the software vendor, Workday, and a federal court has now allowed the case to move forward as a collective action potentially covering a huge number of applicants over 40.

If your business uses AI anywhere in hiring, or in any decision about people, this case is the one to understand. Because the legal principle emerging from it is the same one that ran through the very first article in this series, now with far higher stakes: when you hand a decision to an AI, you don't hand off the responsibility for it.

What Mobley alleged

Workday isn't a niche product. Its human-capital software sits between millions of applicants and the companies they apply to, and its tools can score, sort, rank, and recommend whether a candidate should advance or be rejected. For many applicants, clearing the algorithm is a prerequisite to a human ever considering them.

Mobley's claim, filed in 2023, is that Workday's screening tools discriminate on the basis of age, race, and disability. The mechanism he alleges is the one that makes AI bias so insidious: the system doesn't have to be programmed to discriminate. If it's trained on a company's past hiring patterns, and those patterns favored younger applicants, the AI learns to reproduce that preference and applies it at massive scale, automatically, to every future applicant. Nobody wrote a rule that said "screen out older workers." The system learned it from the data and enforced it silently.

During the case, Workday disclosed that its tools had rejected applications numbering in the billions during the relevant period. That's the scale at which an automated preference operates. A biased human recruiter affects the applicants they personally see. A biased algorithm affects everyone, everywhere the software is deployed, at once.

The ruling that should get your attention

Workday tried to have the case thrown out, and part of its argument is the reason this case matters far beyond hiring.

Judge Rita Lin, in the Northern District of California, issued a mixed ruling in July 2024. She rejected one of the plaintiff's theories but allowed the crucial one to proceed: that Workday could be directly liable as an "agent" of the employers using its software, on the theory that those employers had delegated a traditional hiring function, the decision to advance or reject candidates, to Workday's AI. The federal Equal Employment Opportunity Commission filed a brief supporting that theory.

Then, in May 2025, the court took the next step and conditionally certified a nationwide collective under the Age Discrimination in Employment Act, covering applicants 40 and older who were rejected through the platform going back to 2020. A case that started with one frustrated applicant became a potential action on behalf of a vast class.

Sit with the "agent" theory for a moment, because it's the whole ballgame for any business using AI. The court's logic is that if you delegate a decision to an AI system, the AI is acting as your agent, and you don't escape responsibility for the outcome just because a machine executed it. "The vendor's algorithm did it" is not a defense, any more than "the chatbot said it" saved Air Canada. The decision was yours to make. You chose to let the AI make it. The accountability rides along.

Why this reaches well past hiring

Hiring is the front line here because anti-discrimination law is well developed and the harm is easy to see. But the principle generalizes to any consequential decision a business might automate: who gets approved for credit, whose claim gets flagged, whose file gets prioritized, who gets offered which price.

In each of these, an AI trained on historical data will tend to reproduce whatever patterns are in that data, including the ones you'd never endorse if they were written as an explicit rule. And in each, the emerging legal and regulatory posture is the same: the entity that deployed the system owns its outcomes. The AI's autonomy is not a liability shield. If anything, courts are treating "we didn't know what our own system was doing" as part of the problem, not an excuse.

What to do before you let AI touch a decision about people

Test for disparate impact before and after you deploy. You cannot assume a system is fair because no one designed it to discriminate. Bias enters through the training data, not the intent. Any AI making or shaping decisions about people needs to be tested for whether its outcomes fall unevenly on protected groups, and that testing has to continue after launch, because the system's behavior can drift as data changes.

Keep a human in the loop on consequential decisions. There's a world of difference between AI that helps a person decide and AI that decides on its own. The moment a system can reject a candidate, deny a claim, or close a door with no human review, you've automated not just the work but the accountability, and the accountability comes back to you anyway. Meaningful human review on high-stakes decisions is both the ethical and the legal safeguard.

Interrogate your vendor, and get the answers in writing. If you're buying an AI decision tool, the fact that a vendor built it does not move the liability to them, as Workday's own case is establishing. Ask how the model was trained, what bias testing exists, and what you can audit. If a vendor can't or won't answer, that silence is your answer. You will own the outcomes regardless of who wrote the code.

Preserve an explanation. If your system rejects someone, you should be able to reconstruct why. "The algorithm decided" is not an account you want to give a regulator, a court, or the person who got the 1:30 a.m. rejection. Decisions about people have always demanded a reason. Automating the decision doesn't remove that demand.

The real lesson

We help businesses automate real work, and some of that work involves decisions about people. Done responsibly, AI can make those processes faster and, tested properly, fairer than the inconsistent human judgments they replace. But responsibly is the entire word. It means bias testing, human oversight on the decisions that matter, and a clear-eyed understanding that you, not your vendor, answer for the outcomes.

Derek Mobley's case is still unfolding, and how it ends matters less than the principle it has already established: an AI making decisions on your behalf is doing exactly that, on your behalf. The responsibility never transferred. It was yours the whole time.

A note: this article discusses discrimination and its effects on real people's livelihoods. If any of it resonates with something you're facing, these are matters worth taking to an employment attorney rather than an AI.