The Workday Ruling and What It Means for Your AI Hiring Liability

For two years the open question in AI hiring law was simple to state and hard to answer. If you buy a tool to screen job applicants and that tool rejects people in a way that looks discriminatory, who is on the hook, you or the company that sold you the software? In June 2026 a federal court in the Northern District of California gave the clearest answer yet, and it is the answer that should change how every HR leader thinks about a screening contract.

The case is Mobley v. Workday. The plaintiff, a Black applicant over 40, said he was rejected from more than 100 jobs by employers using Workday's applicant screening, and he sued Workday itself. On June 16, 2026 the court held that Workday can face liability under California civil rights law for AI driven hiring discrimination, including for employers operating outside California. On June 22 the court allowed state discrimination claims to proceed. These rulings sit on top of an earlier finding that the federal age discrimination law reaches applicants, and a nationwide collective covering a staggering pool of rejected applications, all built on one idea: the AI vendor can be treated as the employer's agent.

What the rulings actually held

Strip away the procedural language and three things matter. First, an applicant can bring a discrimination claim against the company that built the screening tool, not only against the employer who used it. Second, the theory that makes that work is agency. The tool was acting on behalf of employers in the hiring function, so it steps into the employer's shoes for liability purposes. Third, the geographic wall came down. A California civil rights claim was allowed to move forward over hiring decisions that reached applicants in other states, which is why this is being read as a nationwide signal rather than a California story.

None of this says Workday discriminated. The case is still being litigated, and a ruling that claims may proceed is not a ruling that the plaintiff wins. What it settles is the shape of the exposure. The door to vendor liability, and through it employer liability, is open.

Why "the vendor is an agent" expands your exposure

Here is the part senior HR leaders should sit with. For years the comfortable assumption was that the vendor carried the technical risk. You licensed a product, you trusted the demo, and if something went wrong the vendor's lawyers handled it. The agent theory takes that comfort away. If the tool acts as your agent in the hiring decision, then its output is your decision, and the discrimination claim attaches to you as the employer who deployed it.

That reframes the whole purchase. You are not buying a feature. You are delegating a regulated employment function to a system, and you keep the responsibility for the outcome. The indemnity clause in the contract may move some dollars around between you and the vendor later. It does not move the applicant's claim, and it does not move the headline.

The practical exposure for any employer using an AI hiring tool

If you screen, rank, match, or filter applicants with software that uses machine learning, this reaches you. The exposure is not limited to companies that bought Workday. The theory is about how the tool functions in the hiring decision, so any resume scorer, video interview analyzer, chatbot screener, or automated knockout question set is in the same conceptual bucket.

Two failure modes are worth naming. The first is the clean audit that gets treated as the finish line. A bias test certificate tells you the tool cleared one defined check. The June rulings point at the limits of leaning on that test as a courtroom defense, so a passing audit narrows your risk without retiring it. The second is the silent record. Most AI screening leaves a data trail of who was scored, ranked, and cut. In litigation that trail is evidence, and you want it to show a defensible process rather than a black box you never questioned.

A vendor due diligence checklist to run now

You do not need to rip out your stack this quarter. You need a record that shows you did the work of a careful employer. Before you renew or expand any AI hiring tool, get these answers in writing. This is a checklist you can build and pressure test with Claude as a working draft, then take to the vendor.

• What does the model actually screen on. Ask for the inputs and the features that drive a score. If the vendor cannot or will not say, that silence is your answer.
• What data was it validated against, and does that population look like your applicant pool. A tool tuned on a different workforce can miss your reality entirely.
• What did the bias testing measure, and what did it not measure. The gaps tell you where your residual risk lives.
• What does the human in the loop actually decide. A reviewer who rubber stamps a ranked list is not meaningful oversight. Define a real decision point and document it.
• What records does the tool keep, and can you retrieve them. You want adverse impact data and decision logs you can produce, not a vendor portal you cannot export.
• What does the contract say about cooperation in a claim. Indemnity is one thing. Getting the vendor's data and testimony when you need it is another.

Run this on every tool, save the answers, and re run it at renewal. That paper trail is the cheapest insurance you can buy against the agent theory, because it converts "we trusted the software" into "we questioned the software and kept the record."

What AI does not replace

It is tempting to read all this as a reason to add more software, another auditing layer, another dashboard. That is backwards. The rulings reward employer judgment, not more automation. AI can draft your vendor questionnaire, summarize a model card, and flag where a screening pattern looks off. It cannot decide what fairness means for your roles, it cannot own the legal call, and it cannot be the human in the loop. That is the work that stays with you, and it is exactly the senior judgment these decisions are pushing back to the employer.

The throughline matches a point we keep making in our hiring briefings. Passing an audit is not the same as being fair, and now a court is treating the gap between the two as something the employer owns.

If you lead hiring and want to build the full operating system for this, the Leveraged HR Professional course walks senior HR leaders through running these tools with Claude without handing over the judgment that keeps you defensible. For the related context, see our briefings on the EEOC enforcement plan for AI hiring tools, why a passed audit is not the same as a fair tool, and what the Colorado AI hiring law asks of HR. Not sure where to start, the two minute course quiz points you to the right one.

Frequently Asked Questions

Does the Workday ruling mean only Workday customers are exposed?

No. The June 2026 rulings turn on how an AI tool functions as the employer's agent in the hiring decision, not on the vendor's brand. Any employer using a machine learning tool to screen, rank, or filter applicants should read the agent theory as reaching them, because the same logic applies to resume scorers, video interview analyzers, and automated knockout screens.

If our AI hiring tool passed a bias audit, are we protected?

A passed audit shows you ran one defined test, which lowers risk but does not retire a discrimination claim. The court in Mobley v. Workday has signaled limits on how far a third party bias test goes as a defense. Treat the audit as a floor, keep your own adverse impact data, and document a real human decision point in the hiring process.

What is the first thing HR should do after these rulings?

Before your next renewal, get written answers from each AI hiring vendor on what the model screens on, what data it was validated against, what the bias test did and did not measure, and what the human reviewer actually decides. Save those answers and re run the questions at renewal, so your file shows a careful employer rather than a black box you never questioned.