Workday Ruling Says Where You Run Your Hiring Model Can Decide Which Law Applies

Workday once told the court that “1.1 billion applications were rejected using Workday” during the period this lawsuit covers [4]. The company offered that number in 2025 as a reason not to notify potential plaintiffs. It didn’t work then, and the size of that funnel is the backdrop for what happened Monday.

On June 22, U.S. District Judge Rita Lin ruled that California’s main anti-discrimination statute can reach Workday’s screening of applicants who never lived or applied for jobs in California [1]. Her reason wasn’t where the applicants were. It was where the software allegedly runs.

What the judge actually decided

The case is Mobley v. Workday, filed in 2023 in the Northern District of California [1]. Derek Mobley, Jill Hughes, Sheilah Johnson-Rocha and FaithLinh Rowe allege that Workday’s algorithm-based screening tools discriminated against them by race, age and disability [1]. Reuters describes it as the first proposed class action to broadly target the algorithmic decision-making in AI screening software [2].

Some history. In May 2025, Judge Lin preliminarily certified a nationwide collective under the federal Age Discrimination in Employment Act, covering applicants 40 and over who used Workday’s platform from September 24, 2020, onward [4]. That order described two tools, including Candidate Skills Match, which grades an applicant’s fit for a posting as “strong,” “good,” “fair” or “low” [4]. In March 2026 she threw out the plaintiffs’ new claims under California’s Fair Employment and Housing Act (FEHA), because the complaint never said what unlawful conduct happened in California [3]. She cited an earlier case holding that the location of a company’s main office wasn’t enough on its own [3].

So the plaintiffs filed a Third Amended Complaint saying exactly that. It alleges that Workday’s screening systems, “including its artificial intelligence and machine-learning tools used to evaluate, rank, and reject applicants,” are “designed, developed, maintained, and controlled” from its headquarters in Pleasanton, and that its training and datasets are run from California for out-of-state customers too [1].

That was enough. The court held the complaint adequately alleges that Workday, “through its algorithmic decision-making tools,” materially participated in the rejections from California [1]. The order rests on four moves:

• The vendor is liable on its own. Workday argued it could only be liable if its employer-customers were. The judge rejected that, citing the California Supreme Court’s 2023 decision in Raines, which held that a business acting as an employer’s agent can be directly liable under the FEHA when it has at least five employees and carries out FEHA-regulated activities for the employer [1, 5].
• It’s not extraterritorial. Applying the FEHA to “a company headquartered in California for its own California-based conduct” doesn’t raise serious constitutional concerns, the judge wrote, and conduct inside the state isn’t extraterritorial “regardless of where an aggrieved worker resides and is employed” [1].
• The disability theory survives. Hughes now alleges that hiring tools can rely on “proxy indicators of illness or health-related limitations,” such as medical leave or “patterns consistent with treatment and recovery” [1]. Workday didn’t contest whether that states an ADA claim, only whether it was allowed in; the court said it was [1].
• Two additions are out. A new race-based disparate impact claim on behalf of Rowe, who is Asian American, was dismissed because the plaintiffs never asked permission to add it, and so was a theory that Workday is liable as an employer for its own hiring [1]. No further leave to amend was granted [1].

Workday’s answer, through Reuters, is that the claims are false and its AI recruiting tools don’t make hiring decisions “in California or anywhere else” [2]. “Our technology looks only at job qualifications, not protected traits like race, age, or disability,” the company said [2].

What it means if you build things

Start with the caveat, because it matters. This is a pleading-stage order. The judge took the complaint’s allegations as true and asked only whether they state a claim [1]. Nobody has proven that Workday’s models discriminate.

Still, two pieces of reasoning travel well beyond Workday.

Your infrastructure map may be your legal map. The hook for California law was the allegation that the tools were developed and operated from California [1]. Footnote 3 is the line to read: the court “does not reach” whether development in California, standing alone, would be enough [1]. So a team that trains in San Francisco and serves from a data center in Virginia has an open question, not an answer. A team that does both in California has this order to contend with. I don’t think anyone picks a cloud region to manage discrimination exposure, and I’m not suggesting you start. But if your ops docs say “our scoring service is controlled from our California office,” that sentence now has a legal reading.

The vendor is a defendant, not a bystander. Raines is the bigger deal for small companies. Its threshold is five employees [5]. That’s most seed-stage startups. The California Supreme Court’s reasoning was that an agent doing “a narrowly defined task for multiple clients over the course of several years” can more likely bear the cost of counsel to check its methods [5]. That logic was written about a medical screening company in 2023. Applied to a ranking API, it means the eight-person startup selling resume scoring is in the same legal category as a vendor whose platform logged 1.1 billion rejections by its own count [4], and the opinion’s assumption that it can afford lawyers doesn’t scale down.

Judge Lin also repeated a point from Raines that should land with anyone selling into HR: the employer “may well” have no knowledge of the agent’s discriminatory practices, and a vendor serving many employers opens the door to “systemic industry-wide” problems [1]. Translated: your customers’ ignorance of how your model works isn’t your shield. It’s part of the court’s reason for coming after you.

And the proxy theory is concrete enough to test against. If your features include tenure gaps, leave history or anything that tracks medical treatment, you should know how they move outcomes for people with disabilities before a plaintiff’s expert does [1].

Duane Morris’s defense-side summary reads the ruling the same way: direct liability under Raines, not liability borrowed from the customers [6]. What it doesn’t settle is the larger question of what happens when California law follows a California vendor into hiring decisions in other states.

My policy view, for what it’s worth: this is how AI hiring rules are getting made right now, one motion to dismiss at a time. A company can’t comply with a pleading-stage order. It can only read the footnotes and guess. That favors whoever can afford to litigate, which is the opposite of what the Raines court assumed about agents.

What I’m watching

• June 29. Workday has to file a proposed order with a version of the complaint that strikes the dismissed allegations [1]. That will show exactly what’s left in the case.
• Choice of law. The judge said how a choice-of-law analysis would apply is “better performed on a fuller factual record,” such as at class certification or summary judgment [1]. That’s where another state’s law could push back.
• The constitutional objection. Workday’s lawyers mentioned the Commerce, Full Faith and Credit and Due Process clauses at the hearing, but only in passing, so the order didn’t address them [1]. If those arguments come back fully briefed, they could matter more than anything decided Monday.