Who Owns It When Your AI Hiring Tool Gets Sued? 4 Critical Points to Accountability

A few recent court cases raise a question that almost nobody inside companies using AI hiring tools has asked before the filings arrived. Who was responsible for what the software was doing?

In one case, Workday disclosed in court filings last May that its software had rejected 1.1 billion job applications during the period at issue in a class-action lawsuit. Mobley v. Workday was certified as a nationwide collective action covering every applicant over 40.

A second lawsuit filed in January 2026 against Eightfold AI made the pattern harder to ignore. Erin Kistler, the named plaintiff, alleged that Eightfold scraped data on more than a billion workers and scored them on a zero-to-five scale. Low-ranked applicants were tossed before any human ever saw their materials, and Eightfold’s clients included Microsoft, PayPal, Morgan Stanley, Starbucks, and Chevron. Kistler told reporters, “I think I deserve to know what’s being collected about me and shared with employers.”

Is the AI Hiring Tool Responsible?

As these patterns unfold inside more companies, we’re starting to see the answer on who’s responsible for what the software does: At most companies, it’s nobody. Vendors didn’t own it. HR didn’t own it. Legal never saw it.

These aren’t fringe cases anymore. SHRM’s 2026 report put AI adoption for HR tasks at 43%, nearly double the 26% recorded just a year earlier. That is the fastest rollout of any HR technology in the past decade. But only 49% of organizations using or piloting these tools have any policies around them, and just 25% of those consider their policies clear enough to hold up.

Tools arrived before rules did. And the gap keeps widening.

Vendor Contracts Won’t Save You

Every employment attorney I’ve talked to about this keeps coming back to the vendor agreements. Jones Walker and Gouchev Law analyzed AI vendor contracts and found that 88% cap their own liability, often limiting damages to a single month’s subscription fee.

Only a third provide any indemnification for third-party claims. When a class-action hits, the contract says the employer takes the full loss. Think about that in the context of Mobley, where the alleged harm spans 1.1 billion application decisions. Your vendor’s monthly fee isn’t covering that.

Keith Sonderling, a former EEOC Commissioner, put it this way. “If the AI in the HR system is not properly designed or used, it can scale discrimination to the likes that we have never seen before.”

That’s not an exaggeration, and Stanford researchers proved the point in October 2025 when they tested AI resume-screening tools. Older male candidates received higher ratings than both female and younger candidates, even though every resume came from identical data.

But Sonderling also said something worth sitting with. “When you talk about the black box of the algorithm, what about the black box of the human brain?” He has a point, and a well-audited AI system creates a better paper trail than any hiring manager’s gut feeling ever did.

AI itself isn’t the problem. You cannot defend a system you are not testing.

Nobody Is Watching

Colorado’s AI Act took effect June 30, 2026. Employers need to complete written impact assessments before deploying high-risk AI, run a documented risk-management program, and tell candidates when AI plays a role in hiring. Illinois already passed its own law, effective January 2026. California demands four-year data retention and trained human oversight with real override capability. Three states, three different sets of rules. If your HR team hasn’t started preparing for any of them, you aren’t alone.

And 57% of HR professionals in states with these laws on the books don’t know they exist. SHRM reported that number. If you work in one of those states, read it again.

New York City tried to get out ahead of this with Local Law 144, requiring bias audits for automated hiring tools. NYC’s Comptroller audited enforcement from July 2023 through June 2025, and the results were ugly. Seventy-five percent of test calls to the city’s 311 hotline about AI hiring were misrouted and never reached the right agency. Enforcement staff surveyed 32 companies and found one case of non-compliance. But the Comptroller’s own auditors, looking at those same companies, identified at least 17 potential violations. On paper, the law existed, but in practice, it was barely functioning. If that’s the best-case example of AI hiring regulation in this country, the rest of it is flying blind.

Who Should Own This

Udbhav Ganjoo, Head of HR for Global Operations at Viatris, was direct about where accountability should land. “Ownership of the human impact of AI cannot sit anywhere but HR,” he said. Pinnacle’s 2026 survey of enterprise CHROs told a different story, though. Only 21% are closely involved in AI decisions. Ninety-two percent said they participate at “some level,” which is a polite way of saying most are being briefed, not deciding.

Betsy Summers at Forrester told me she’d push back on the pace of adoption. “I would slow down,” she said. “I don’t think that speed to value is the right selling point here.”

And Mercer’s data backs up why that matters. Fewer than one in four CEOs have said anything publicly about how AI will affect jobs. Fewer than one in five employees have heard from their managers about it. People buying these tools aren’t talking to the people affected by them, and that silence has a cost when a lawsuit drops.

Maggie Ruvoldt, CHRO at Learn Behavioural, had a practical line. “All the yeses can automatically move forward, but all the no’s have to be reviewed by somebody.”

Most companies aren’t hitting even that bar. Gartner found that only 26% of candidates trust AI to evaluate them fairly, while 52% believe their applications are already being screened by it. People know these systems exist, but they don’t believe anyone is watching.

Rebecca Wettemann at Valoir said vendors need to explain how their algorithms work “to your satisfaction, not to their satisfaction.” If the person at your company who approved the tool can’t explain what it does in plain language, you don’t have accountability. You have a purchase order.

A 4-Point Accountability Checklist

Before the next AI hiring lawsuit becomes your problem, answer these four questions in writing. If you can’t answer one of them without asking somebody else, you’ve found a gap.

1. Who signed the vendor contract, and have they read the liability cap? Pull the agreement from your procurement team. Find the indemnification clause and the damages cap. If the person who signed can’t tell you what those clauses actually say in plain language, your company is exposed and doesn’t know it. The Jones Walker data is suggesting that you probably aren’t covered for the kind of class action damages that can show up in cases like Mobley.

2. Who reviews the rejections before they go out? This is essentially the Ruvoldt test. Every single candidate the system rejects should have a human checking the decision before it becomes finalized. If your workflow doesn’t have a human in the loop, then you are rubber-stamping an algorithmic decision with a human signature; that signature and ultimate accountability is yours.

3. When was the tool last audited, and by whom? You’re going to need a calendar entry for bias testing at least twice a year. The tester shouldn’t be the vendor who built the tool. It needs to be third-party and unbiased. If your last audit happened during implementation and you haven’t had anything since, assume that you have a current problem because models drift and training data ages. The Stanford study is not a one-time finding.

4. Do your candidates know AI is involved? Some states require disclosure. If your candidate-facing and communication don’t mention AI anywhere, then you’re out of compliance in more states every quarter as more and more follow this cadence. Fix that before a candidate asks or a relative regulator does. Keep this list somewhere your procurement team, your HR leaders, and your legal counsel can all see. That’s what accountability looks like in practice, and it’s a lot cheaper than what comes next when you don’t have it.