Algorithmic Rejection

My friend K landed a job seven months after being laid off just a few weeks after she returned from maternity leave. She's been meticulously keeping track of the rejection letters and shared this story with me about Derek Mobley. He detected  the pattern when he applied for more than 100 jobs through the Workday platform and was rejected from each one, sometimes within hours or minutes of submitting his application. Exactly K's experiecnce and that for countless other job applicants. To Mobley, the speed was the tell. No human being had read his résumé. No one had weighed his finance and IT degrees against the requirements of the role, considered his experience, or made any judgment about him as a candidate. The algorithm had simply processed him and found him wanting or more precisely, had found in him the markers it had been trained to filter out. Mobley is Black, over 40, and disabled, and his lawsuit alleges that these characteristics were what repeatedly knocked him out of the running. He may or may not be right about the algorithm's rationale for rejection if we even find out what it is.

That said, the case Mobley v. Workday is being watched closely because it tests something the tech industry has preferred to leave ambiguous: whether a software vendor can be held liable for the discriminatory outcomes of tools it built and sold. For years, the standard corporate defense was essentially philosophical. Companies would argue they didn't know why the algorithm rejected a candidate and that the underlying code was a proprietary trade secret, a black box they couldn't open even in litigation. The algorithm did it and so there is nothing to be done about it. A 2024 ruling shifted that calculus considerably, with the court finding that Workday could be held liable as an "agent" of the employer, meaning that tech vendors can no longer simply hand off responsibility to the companies that purchased their tools. The vendor who built the filter is now, at least potentially, as accountable as the employer who deployed it.

What makes this more than a single lawsuit is the scale at which these tools operate. Court filings in the Workday case indicate that its software rejected approximately 1.1 billion applications since September 2020, creating a potential class of hundreds of millions of job seekers. These are not edge cases or statistical anomalies. They are the normal, intended function of a product sold to companies on the promise of efficiency that the expensive, time-consuming work of evaluating human beings could be automated away. A 2024 Gallup survey found that 93% of Fortune 500 Chief Human Resource Officers are integrating AI into business practices, yet only about a third of employees knew their employer uses AI tools in hiring. While the screening is everywhere, the awareness of it is not.

The deeper problem is one the lawsuit cannot fully solve even if it succeeds. An algorithm trained on historical hiring data does not invent bias from scratch because it inherits and accelerates the biases already present in the decisions human beings made before it existed. If the data used to train the algorithm reflects biased hiring patterns that has been favoring certain genders, races, ages, or schools,  the AI will replicate those patterns at scale. Winning a civil rights case against Workday would be meaningful and overdue. It would not, by itself, answer the harder question underneath: what happens when the discrimination is not a glitch in the system, but a faithful reproduction of it. The midnight rejection email is the symptom. The bias it encoded was already there, in the decades of human hiring decisions that taught the machine what a desirable candidate looks like.