Navigating the Ethics of AI Interviews: Compliance and Candidate Trust

EEOC, the EU AI Act, and candidate consent: a practical playbook for teams rolling out voice AI that won't land you in a lawsuit.

Voice AI for interviews is moving faster than the regulatory perimeter. That's an opportunity and a trap. The teams who think about this first pay a small upfront tax. The teams who don't pay a very large backfit tax later.

The four non-negotiables

Disclose before the first question

"You're about to speak with an AI interviewer." Not in a PDF attached to the invite. On the call, in the first sentence, with a confirmation prompt.

Offer a human path

Candidates who decline AI interviews can't be silently penalized. A human-run alternative must exist and must not be obviously slower or worse.

Audit for disparate impact

Pull pass-rates by demographic category quarterly. If a protected group passes at materially lower rates, the model is a legal problem regardless of intent.

Retain explainability

For every reject, you should be able to show a human-readable reason in under 30 seconds. If your system can't do this, it's not ready for production.

Compliance isn't a feature list. It's a design constraint. Bolt it on at the end and you'll rebuild half the product.

What candidates actually ask

In our data, the top three candidate concerns aren't technical. They're human: Will I get feedback? Who sees my recording? Can I request a human reviewer? Build your flow to answer those three in the first screen and trust goes up — measurable in completion rate.

The hard line:: If the only reason you're using AI interviews is because they're faster, you'll lose a lawsuit eventually. Use them because they're more consistent — and prove it.