PreviewYou're previewing CHRO & People Compass. Sign in to use it.Sign in
← All lessons
L2 · Fair and compliant AI in hiring and employee decisions9 min read

Bias audits and the questions to ask HR-tech vendors

Most HR AI comes from vendors. Your leverage is before you sign and at every renewal. Here is what to ask and check.

What a bias test looks like

In plain terms, a bias test compares outcomes across groups: of the people the tool assessed, what share in each group did it advance or score highly? Large gaps between groups are a warning sign that needs an explanation linked to the job — or a fix. Tests should cover sex, race/ethnicity and their combinations, plus age and disability where data allows, and be repeated after model changes. In New York City, a yearly independent bias audit with a published summary is a legal requirement for automated employment decision tools.

Ten questions for vendors

  1. What exactly does the AI do — predict, draft or act — and at which hiring step?
  2. Which data does it use, and which features drive its scores? Can you explain them in plain words?
  3. What bias testing have you done, on what data, and when? Can we see the results?
  4. Will you support our own audit on our data, including a yearly NYC-style bias audit?
  5. How will you notify us of model changes, and will you re-test after them?
  6. Can candidates request a human review or an alternative process?
  7. Is our candidate and employee data used to train your models? Can we opt out?
  8. Where is data stored, for how long, and how is access controlled?
  9. How will you meet EU AI Act provider duties for high-risk systems by December 2027?
  10. Have you faced complaints, regulatory inquiries or lawsuits over the product?

Why vendor diligence matters now

In Mobley v. Workday, a U.S. federal court in May 2025 preliminarily certified a nationwide collective of applicants aged 40 and over who alleged that the vendor's AI-based recommendation system discriminated by age — reportedly the first such certification for AI hiring tools. Employers using the same tools should expect questions too.

Put it in the contract

Bias test results on request, audit cooperation, notice of material model changes, no training on your data without consent, security standards, and the right to suspend a tool that shows adverse impact.

What this means for the CHRO

Make vendor AI questions a standard part of HR-tech procurement, owned jointly by HR, legal, procurement and security.

No bias evidence, no explanation of features, no audit support → no contract. And schedule a yearly re-test in your HR-tech calendar, the same way you renew licenses.

Keep the answers on file: they are your evidence if a candidate, regulator or court asks how you chose and checked a tool. Re-ask the questions whenever a vendor releases a major new AI feature, and assign one owner in HR to chase overdue re-tests.

Real example

On May 16, 2025, a U.S. federal court preliminarily certified a nationwide collective action under the Age Discrimination in Employment Act in Mobley v. Workday, on behalf of applicants aged 40 and over who alleged that Workday's AI-based applicant recommendation system discriminated against them — which lawyers described as the first such certification over AI hiring tools. Source: Davis Wright Tremaine — Federal court allows age-discrimination collective action over AI hiring tools (Mobley v. Workday, May 2025) ↗

Use it

Quick check · 3 questions

  1. 1.What does a basic bias test compare?

  2. 2.When should a hiring tool be re-tested?

  3. 3.Which contract clause best protects you?

0 of 3 answered

Sources

Last reviewed 2026-10-01.