[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fLgs2uoFDSY_dlTBOjq1xHEyNItJ6sTOaDeJD9R_720o":3,"$fm6Ky0r9HHO8PTXKBgXJeU4fCLm4z6RKRMXFRXG3f560":28},{"success":4,"data":5},true,{"id":6,"slug":7,"title":8,"excerpt":9,"content":10,"category":11,"tags":12,"author":20,"cover_image_url":21,"reading_time_minutes":22,"is_published":4,"published_at":23,"created_at":24,"updated_at":24,"author_avatar":21,"is_featured":25,"meta_title":26,"meta_description":27,"meta_keywords":21},"6e431f94-fecc-44be-aff1-0fbbcb99c511","mobley-v-workday-ai-hiring-vendor-liability","The Workday Lawsuit Isn't About Workday: What Three Years of Mobley v. Workday Rulings Mean for Every AI Hiring Vendor","Mobley v. Workday has spent three years and six rulings deciding whether an AI hiring vendor — not just the employer using its tool — can be sued directly for discrimination. Here's what's actually settled, what's still moving, and what it means for anyone building or buying algorithmic screening tools.","\u003Cp>Most compliance news about AI hiring is written for the company doing the hiring. \u003Cem>Mobley v. Workday, Inc.\u003C\u002Fem> is different. It is the case that decided whether the company that built the screening software can be sued directly — not the employer that used it, the vendor. Three years and a half-dozen rulings in, the U.S. District Court for the Northern District of California has now said, repeatedly and with increasing specificity: yes. That answer matters to every company in the business of building algorithmic hiring tools, NiceHire included, whether or not it ever appears in a US courtroom.\u003C\u002Fp>\n\n\u003Cp>We build screening software. So we read this case the way a vendor reads it, not the way a law firm's client alert reads it — and that means being specific about which parts are settled, which are still moving, and which conclusions people are drawing that the rulings don't actually support.\u003C\u002Fp>\n\n\u003Ch2>The case, briefly\u003C\u002Fh2>\n\n\u003Cp>Derek Mobley, a Black man over 40 with documented anxiety and depression, sued Workday in 2023 after being rejected from more than 100 jobs at companies that use Workday's applicant-screening tools. His claim was not against the employers — it was against Workday itself, on the theory that its AI recommendation and scoring engine was the thing that actually screened him out, and that Workday should answer for that under federal anti-discrimination law the same way an employer would.\u003C\u002Fp>\n\n\u003Cp>Workday's first line of defense was structural: Title VII, the ADEA, and the ADA prohibit discrimination by an \"employer\" or \"employment agency\" — and Workday is neither of those things to Mobley. It never hired him, never employed him, never placed him. It licenses software to the companies that did.\u003C\u002Fp>\n\n\u003Cp>On July 12, 2024, Judge Rita Lin rejected the \"employment agency\" framing but let a narrower one through: those statutes also reach an employer's \u003Cstrong>agent\u003C\u002Fstrong>, and a vendor performing a function an employer would otherwise perform itself — screening, scoring, recommending, rejecting — can be an agent for that purpose. That single ruling is why this case outlived its first motion to dismiss and became the one every AI-hiring vendor's counsel now cites by name.\u003C\u002Fp>\n\n\u003Ch2>What's happened since — the part most coverage skips\u003C\u002Fh2>\n\n\u003Cp>The \"agent theory\" ruling is the famous one, but it's the two years of follow-on rulings that show what it actually does in practice:\u003C\u002Fp>\n\n\u003Cul>\n\u003Cli>\u003Cstrong>May 16, 2025 — nationwide collective certification.\u003C\u002Fstrong> Lin certified an ADEA collective covering everyone 40 or older who applied through Workday's platform since September 24, 2020 and was denied a recommendation — not a single plaintiff's grievance, a nationwide opt-in class. Workday's own court filings put the number of applications rejected through its system in that window at 1.1 billion. The court later approved a plan (December 2, 2025) to notify prospective members so they can opt in.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>March 6, 2026 — the applicant-coverage fight.\u003C\u002Fstrong> Workday argued the ADEA doesn't cover applicants at all, only employees — an argument aimed squarely at unwinding the collective action a year after it was certified. Lin rejected it, pointing to Supreme Court precedent Workday's reading couldn't get around. In the same order she dismissed an individual plaintiff's disability claim and some California state-law claims, with leave to fix and refile by March 27.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>May 2026 — a privilege ruling that cuts both ways.\u003C\u002Fstrong> Lin held that some of Workday's internal AI bias-testing data can be protected from discovery under attorney-client privilege, when the testing was done at counsel's direction for legal-advice purposes. Read generously, this rewards companies for actually testing their models for disparate impact under privilege rather than not testing at all. Read less generously, it means the testing a court might otherwise use to establish (or disprove) bias can be shielded from the people asking the question. Both readings are live; the ruling is narrow and fact-specific, not a blanket shield.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>June 22–24, 2026 — the nexus ruling.\u003C\u002Fstrong> On the amended complaint, the court again largely denied Workday's motion to dismiss, this time allowing California Fair Employment and Housing Act (FEHA) claims to proceed even for applicants who were screened for jobs outside California. The reasoning: Workday designs, develops, and operates its screening tools from its California headquarters, so California's law can reach that conduct regardless of where the rejected applicant sat or which state's job they applied for.\u003C\u002Fli>\n\u003C\u002Ful>\n\n\u003Cp>That last point deserves the most attention, and gets the least. It is not really a ruling about Workday's location — it's a template. Any vendor headquartered somewhere with an assertive anti-discrimination or AI statute could, on the same logic, find that state's law following its product everywhere the product runs, not just where its customers or their applicants happen to be.\u003C\u002Fp>\n\n\u003Ch2>What is settled, what isn't, and what we think\u003C\u002Fh2>\n\n\u003Cp>\u003Cstrong>Settled, for now:\u003C\u002Fstrong> an AI vendor performing employer-like screening functions can be sued directly as an \"agent\" under Title VII, the ADA, and the ADEA (July 2024, reaffirmed through subsequent rulings); the ADEA covers applicants, not just employees (March 2026); a state's employment-discrimination law can reach a vendor's conduct based on where the vendor built and runs the tool, not only where the affected applicant or employer sits (June 2026).\u003C\u002Fp>\n\n\u003Cp>\u003Cstrong>Genuinely unsettled:\u003C\u002Fstrong> whether Mobley's plaintiffs can ultimately prove disparate impact or disparate treatment on the merits — every ruling to date is about whether the case can proceed, not whether Workday actually discriminated; how far the privilege protection for bias-testing data extends beyond this case's specific facts; and whether other circuits will adopt the Ninth Circuit district court's \"agent\" reasoning at all, since no appellate court has yet weighed in.\u003C\u002Fp>\n\n\u003Cp>\u003Cstrong>Our reading, labeled as ours:\u003C\u002Fstrong> the direction of travel across five separate rulings has been consistently toward more vendor exposure, not less, and Workday has not won a motion that ends any of the core theories — only narrowed some individual claims. A vendor's practical assumption should be that \"we're just the software\" stopped being a reliable defense somewhere around July 2024, and that assumption should shape how screening tools are built, not just how they're marketed.\u003C\u002Fp>\n\n\u003Ch2>What this means if you build these tools — or buy them\u003C\u002Fh2>\n\n\u003Cp>For a vendor, the case argues for exactly the boring things compliance people have been saying for years, now with a concrete reason to actually do them: keep records of what a model was asked to evaluate and on what basis; don't let automation make a final rejection decision without a human able to override it; document bias testing formally rather than informally, because informal testing gets none of the privilege protection and formal testing might; and treat \"the customer configured it that way\" as a weaker defense than it used to be, since the agent theory reaches conduct the vendor itself designed and operates.\u003C\u002Fp>\n\n\u003Cp>For a buyer evaluating vendors, Mobley is a reason to ask harder questions in procurement, not to assume the vendor's own terms of service settle the liability question: what happens to a candidate the model scores below threshold — is a rejection automatic, or does a human see it first? Is every candidate for a role asked the same questions, or does the process vary in ways that are hard to defend as consistent? Is there a record of what the tool actually evaluated, or only a final score?\u003C\u002Fp>\n\n\u003Cp>Those are the questions our own product answers, and it's worth being precise about how, because precision is the whole point of this piece. NiceHire's screening stages support automatic or manual progression, but auto-reject below the passing threshold ships off by default — a candidate who scores below the bar routes to pending review for a human decision, not an automatic rejection, unless an organization opts a stage into auto-reject. Every applicant to a role is asked from the same defined question set in the same order. A completed AI screening stores a structured evaluation — technical, communication, and cultural-fit scores, an overall recommendation, and recorded strengths and gaps — against the application, not just a pass\u002Ffail flag. Every AI screening greeting identifies the interviewer as an AI, in every language NiceHire ships, on first attempt and reconnect alike. Screening interviews record a transcript against the application, with two honest limits: it's whatever the underlying provider returns, so a session that ends without one is stored without one, and re-running a screening replaces rather than versions the prior attempt's transcript.\u003C\u002Fp>\n\n\u003Cp>We're stating those plainly rather than bundling them into a single reassuring phrase, because a bundled claim is exactly the kind of thing a plaintiff's exhibit list turns into evidence of overpromising. None of this is a legal defense, and we're not claiming it is one — Mobley is a case about vendor conduct, not vendor marketing copy, and no design choice substitutes for actually being tested in a case like this one. What these choices are is a bet about which direction hiring-AI liability is heading, made before a court told us the answer, because the alternative — building an opaque, fully automated rejection pipeline and hoping \"we're just the software\" holds up — is the bet Mobley is actively testing in a California courtroom right now.\u003C\u002Fp>\n\n\u003Ch2>The geography, and why it doesn't make this someone else's problem\u003C\u002Fh2>\n\n\u003Cp>NiceHire is built in Hong Kong by Vectech Limited, and our home markets are Hong Kong and Japan — not the jurisdiction Mobley is being litigated in, and the ADEA\u002FTitle VII agent theory doesn't reach across the Pacific by itself. It would be convenient to file this under \"US employment law, not our problem,\" and inaccurate to do so. We've written before about how \u003Ca href=\"https:\u002F\u002Fwww.nicehire.ai\u002Fblog\u002Fhk-pcpd-model-ai-framework-hiring\">Hong Kong's PCPD\u003C\u002Fa> has been running its own AI compliance checks against employers under the PDPO, naming applicant assessment as a higher-risk use, and about how the EU AI Act's high-risk regime for recruitment AI — deferred to December 2027, but not abandoned — still requires human oversight and documentation regardless of where the vendor sits. Three different legal systems, three different theories of liability, converging on the same handful of practical expectations: a human somewhere in the loop, a record of what the system was told to evaluate, and disclosure to the person on the other end of the interview. Mobley is the sharpest test of that convergence so far, because it's the first of the three to ask whether the company that wrote the code, rather than the company that used it, can be made to answer for what the code did.\u003C\u002Fp>","Compliance",[13,14,15,16,17,18,19],"AI hiring","employment law","Mobley v. Workday","algorithmic bias","HR compliance","vendor liability","AI Act","NiceHire Team",null,9,"2026-09-13T00:19:17.41+00:00","2026-09-13T00:19:17.563692+00:00",false,"Mobley v. Workday: What AI Hiring Vendors Need to Know","A vendor's read on three years of Mobley v. Workday rulings — the agent theory, the California nexus ruling, and what it means for anyone building or buying AI screening tools.",{"success":4,"data":29},{"posts":30,"count":72,"hasMore":4},[31,33,44,60],{"id":6,"slug":7,"title":8,"excerpt":9,"category":11,"tags":32,"author":20,"cover_image_url":21,"reading_time_minutes":22,"published_at":23},[13,14,15,16,17,18,19],{"id":34,"slug":35,"title":36,"excerpt":37,"category":11,"tags":38,"author":20,"cover_image_url":21,"reading_time_minutes":22,"published_at":43},"791a5130-dacb-43bc-b7f5-1cc93b103ea1","illinois-enforces-colorado-retreats-washington-intervenes-the-2026-whiplash-in-ai-hiring-law","Illinois Enforces, Colorado Retreats, Washington Intervenes: The 2026 Whiplash in AI Hiring Law","In 2026, US AI-hiring law splintered four ways at once: Illinois enforces on schedule, Colorado's law was enjoined then rewritten and delayed to 2027, NYC's own auditor called its bias-audit law's enforcement \"ineffective,\" and a federal executive order is trying to preempt all of it. Here is what actually holds constant.",[13,17,39,40,41,42,14],"AI regulation","Illinois HB 3773","Colorado AI Act","Local Law 144","2026-09-07T00:20:34.861+00:00",{"id":45,"slug":46,"title":47,"excerpt":48,"category":11,"tags":49,"author":20,"cover_image_url":21,"reading_time_minutes":58,"published_at":59},"5e980715-9c25-4b9a-a6f3-dc701fdbfc90","hong-kongs-reference-check-net-now-reaches-110000-financial-workers-what-phase-3a-means-for-hiring-timelines","Hong Kong's Reference-Check Net Now Reaches 110,000 Financial Workers: What Phase 3A Means for Hiring Timelines","Since 1 July 2026, Hong Kong's Mandatory Reference Checking regime crosses the wall between banks and insurers for the first time, reaching roughly 110,000 employees. Heres what Phase 3A actually requires, what the early data shows, and what a month-long reference-check turnaround does to hiring timelines for regulated roles.",[50,51,52,53,54,55,56,57],"Hong Kong","financial services","background checks","HKMA","compliance","hiring pipeline","recruitment","insurance",8,"2026-09-03T00:20:22.04+00:00",{"id":61,"slug":62,"title":63,"excerpt":64,"category":11,"tags":65,"author":20,"cover_image_url":21,"reading_time_minutes":70,"published_at":71},"4d06101f-bc96-4090-a72b-9798e3018e04","colorado-ai-hiring-law-comment-deadline","Colorado Rewrote Its AI Hiring Law — Public Comments Close in Two Days","Colorado replaced its 2024 AI Act with SB 26-189, a narrower notice-disclosure-correction-review framework covering hiring decisions. Draft rules are open for public comment until 4 September 2026, with a hearing 26 October, ahead of the 1 January 2027 effective date.",[13,66,39,54,67,68,69],"Colorado","ADMT","US employment law","hiring technology",7,"2026-09-02T00:17:55.903+00:00",17]