Your Resume Got Rejected by an Agent

Introduction

Workday's own court filing put the number on the record: its software rejected 1.1 billion job applications during the period a discrimination case covers. On June 22, 2026, a federal judge ruled Workday is an "employment agent" under California law, which means it can be sued directly for discrimination its AI carried out. A month before that, a separate lawsuit accused Eightfold AI of quietly scoring more than a billion workers on a 0-to-5 scale, scraped from sources those workers never knew about, before a single human read their resume. Two cases, filed weeks apart, both chipping away at the same legal cover the AI hiring industry has leaned on since this technology shipped: we only sell the software.

What "Agent" Actually Means Here

The defense was always the same. We just sell the software. The employer sets the criteria, the employer makes the call, the employer carries the liability. Workday is a tool, like a spreadsheet, and you don't sue the spreadsheet.

That argument held until July 2024, when U.S. District Judge Rita Lin took a first swing at it. The plaintiff, Derek Mobley, said he'd applied to more than 100 jobs through Workday-powered portals and got rejected every time, sometimes within an hour of hitting submit, one rejection landing at 1:50 in the morning. He's Black, over 40, and has a disability. Lin let the case move forward on a theory that hadn't worked against a software vendor in the Ninth Circuit before: when a company's AI "dispositions" applicants on its own, rejecting some and advancing others before a human gets involved, the vendor is acting as the employer's agent. As she wrote in the order, "Nothing in the language of the federal anti-discrimination statutes or the case law interpreting those statutes distinguishes between delegating functions to an automated agent versus a live human one."

The June 22, 2026 ruling extended that into California's Fair Employment and Housing Act, which is one of the strongest civil rights statutes in the country. Workday tried a geography argument: most class members live outside California, the jobs were elsewhere, so FEHA shouldn't apply. Lin didn't buy it. Because Workday is headquartered in California and its tools were "designed, developed, maintained, and controlled" from there, the screening "originated in California," and the company is directly liable for its own role in FEHA-regulated hiring. For years the live legal question was whether AI hiring tools are biased. Lin's two rulings move it somewhere harder for the industry: whether the company selling those tools can be put on the hook when they are.

The Second Front: A 1970 Law Nobody Applied to AI

While the Workday case attacks the outcome, a companion case attacks the process, and it might be the more dangerous of the two.

Kistler v. Eightfold AI, filed January 20, 2026, makes a deceptively simple argument: Eightfold is a consumer reporting agency, and it's been operating like an illegal one. According to the complaint, citing Eightfold's own marketing, the platform's model draws on "more than 1.5 billion global data points" and "the profiles of more than 1 billion people." It pulls from LinkedIn, GitHub, Crunchbase, resume databases, plus social media, location data, device activity, and cookies. Then it generates a "match score" from 0 to 5 on each applicant, and per the complaint, "lower-ranked candidates are often discarded before a human being ever looks at their application." The named clients include Microsoft, Morgan Stanley, Starbucks, PayPal, and Chevron.

Here's why that's a problem under the law. The Fair Credit Reporting Act, on the books since 1970, says that if you assemble information about someone's "character, general reputation, personal characteristics, or mode of living" and sell it to an employer for a hiring decision, you're a consumer reporting agency. That status comes with rules: register, tell the applicant a report exists, hand them a copy on request, let them dispute errors, send an adverse-action notice before they're rejected on the basis of it. The complaint says Eightfold did none of that. It built dossiers, scored people, fed the scores to employers, and never told the people being scored that any of it was happening.

The reason lawyers are paying attention is the math. FCRA statutory damages run $100 to $1,000 per willful violation, and California's version pushes that to $10,000. As one law firm tracking the case put it, when your database covers a billion profiles, the arithmetic gets uncomfortable fast. And the FCRA theory carries a feature the discrimination cases don't: you don't have to prove bias. You only have to prove the company skipped the disclosures. Eightfold denies it, saying its platform "operates on data intentionally shared by candidates or provided by our customers." The lead attorney on the other side is Jenny Yang, who used to chair the EEOC, which tells you this isn't someone fishing for a quick settlement.

How the Bias Gets In Without Anyone Typing "Age 47"

Workday's response is that its tools look only at job qualifications, never at protected traits. Its filings describe the platform as matching "qualifications listed in a candidate's application" against "qualifications identified by the employer." On its face that sounds clean, but it only describes what the tool is trying to do, not what comes out the other end.

You saw this exact mechanism in health insurance a few months back, where an algorithm used cost as a stand-in for sickness and quietly downgraded Black patients without ever reading race. Hiring works the same way. An algorithm doesn't need to know you're 47 to screen you out for age; it just reads the year you graduated. Your race is legible from the school you attended. A long employment gap can stand in for a disability or for years spent caregiving. These are proxy variables, and they let a system discriminate on a protected trait without ever processing the trait itself. That's the whole ballgame in a disparate-impact case, and it's why "we only look at qualifications" isn't the shield Workday seems to think it is. The qualification data itself is built from decades of human hiring decisions, with all their old biases baked in.

The plaintiffs put a number on it. The amended complaint alleges Workday's screening produced a disparity against African-American applicants running more than 15 standard deviations from what you'd expect if the process were race-neutral, and an even larger gap for women. For context, scientists usually treat five standard deviations as proof a result isn't chance. The case is still in discovery and no court has found Workday's tools actually discriminated. What Lin's June 22 ruling did was let that question reach trial — preserving the FEHA agent-theory and disability claims, while striking a late-added Asian-American race claim from one plaintiff on procedural grounds unrelated to the merits.

Who Benefits

Follow the money and it lands on Workday first. The company pulled in $9.552 billion in revenue for fiscal 2026, built largely on HR platform contracts that increasingly bundle in AI screening. When it bought the AI recruiting firm HiredScore in 2024, the pitch to customers was a 25% increase in recruiter capacity, which is a polite way of saying the AI lets you employ fewer human recruiters. Customers pay Workday for that headcount reduction. Transparency about how the algorithm actually sorts people would put both the revenue and the marketing claim at risk, which makes the opacity a feature the business depends on.

The employers using these tools benefit too, and the benefit is cover. Screening 500 applicants by hand takes a recruiter the better part of a workday per role; at the volume a Fortune 500 company sees, you'd need a much bigger team. The AI makes that problem disappear, and it hands the employer a clean story when someone cries discrimination: we used the recommended vendor, we set neutral criteria, the system did the matching. Lin's agent rulings are dismantling that story in slow motion. The cases will take years, so for now the arrangement keeps running exactly as designed.

There's a counter-incentive worth naming, because it's why these particular cases have teeth. The plaintiffs' side isn't a settlement mill. Outten & Golden and Towards Justice, with a former EEOC chair leading, are running a deliberate campaign to set precedent. If the FCRA theory holds, it could force the entire AI hiring sector to comply with a 55-year-old consumer protection law more or less overnight.

The Regulator Walked Off the Field

Here's the part that turns this from a corporate story into something stranger. The federal agency that was supposed to be policing this had already taken Workday's accusers' side, then reversed course before the win came in.

In April 2024, the EEOC filed an amicus brief in the Workday case arguing that a software vendor supplying a discriminatory hiring tool should be on the hook for it, the same theory Lin ultimately ran with. The brief cleared the commission on a 3-2 party-line vote. Then the administration changed. On January 27, 2025, the EEOC pulled its AI hiring guidance off its website, the guidance that had told employers they couldn't dodge liability by blaming a vendor. An April 2025 executive order went further, directing federal agencies to "deprioritize enforcement of all statutes and regulations to the extent they include disparate-impact liability." Disparate impact is the entire basis of the surviving Workday claims.

Withdrawing guidance doesn't repeal the statute, though. Title VII, the ADEA, the ADA, and the FCRA are all still law. What the rollback did was knock out the federal enforcer, which left private lawsuits as the only mechanism still standing. So the government stepped back from enforcing the rule at the exact moment a courtroom proved the rule still bites, and the agency that filed a brief arguing for this outcome won't be the one collecting on it.

The Bottom Line

The headlines flatten what the June 22 ruling actually did. Lin let the FEHA claim, a disability claim, and the core African-American disparate-impact race claim proceed. She struck a late-added Asian-American race claim from plaintiff Rowe (it exceeded the scope of leave to amend, a procedural issue unrelated to the merits), a defense win buried under the "Workday loses" coverage. The court has not ruled that Workday's tools discriminate. It ruled that Workday can be put on trial for whether they do, which is the door that was supposed to stay shut.

Several states are now writing the same logic into statute. Connecticut's CART Act, signed May 27, 2026 and effective this October, takes Lin's ruling and turns it into black-letter law: using an automated decision tool "is not a defense to a complaint alleging a discriminatory employment practice." Illinois and Colorado have versions of their own moving through. The vendor-as-neutral-tool defense is closing off in courtrooms and legislatures at the same time.

So the open question isn't whether these tools shape who gets hired. The question is whether a system that has rejected more than a billion applications before a human reads them can ever be made to show its work, when the company selling it has every reason not to, and the agency that used to ask just stopped.