180 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
10 new this week Library last updated 30 August 2026
← The incident library
HD-INC-126
Hiring and recruitment technology · United States · 2021 · Opaque AI scoring of job candidates from face and voice

A US company sold employers software that scored job candidates from their recorded face and voice on a video interview, and dropped the facial analysis only after a privacy complaint and an outside audit

By Ellie Harris · Filed EPIC complaint to the FTC 2019; facial analysis dropped January 2021

Alleged: HireVue developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

A US company sold employers software that scored job candidates from their recorded face and voice on a video interview, and dropped the facial analysis only after a privacy complaint and an outside audit

What happened

It was reported that HireVue, a US company, sold employers a video-interview tool that scored job candidates by analysing recordings of them answering questions on camera. The software read a candidate’s facial expressions, along with their words and tone of voice, and turned them into a ranking meant to predict how well the person would do in the job. Employers used it to sift large numbers of applicants, and a candidate could be scored, and screened out, without a person having watched the interview.

It was reported that the approach drew heavy criticism. Researchers argued that reading personality or ability from facial movements rests on weak science, and that such a tool could quietly penalise people who did not express themselves in the expected way, including disabled candidates and those from different cultures. In 2019 the Electronic Privacy Information Center, a US privacy group, complained to the Federal Trade Commission that HireVue’s practice was unfair and deceptive and that candidates had no meaningful way to see or contest how they were judged. In January 2021, after commissioning an outside audit, HireVue said it would stop using facial analysis while continuing to assess a candidate’s language. It said the facial component had added little to the result.

What an auditable version would have shown

A candidate turned down after a video interview had no way to know what the tool had seen in them, what it scored, or why. An auditable version keeps a record for each candidate showing what was measured, the score it gave and the reason for it, and puts that in the candidate’s hands, so a person judged by a machine can look at the judgement and argue with it. It also keeps the wider picture: whether the scores came out evenly across different groups of applicants, so a tool that quietly marks some kinds of people down shows up as a number the employer can see, not a hunch a rejected candidate has no way to prove.

Where the gap was

People were ranked for jobs by a tool that read their face and voice for qualities it may not be able to judge that way at all, and neither the candidate nor, often, the employer could see how the score was reached. A ConductRecord keeps each candidate’s assessment, what the system measured and the basis for the score, so the judgement can be shown to the person and questioned. A MetricRecord tracks how the scores come out across groups of candidates, so a pattern that disadvantages some is a number the employer holds. HireVue had to bring in an outside audit to find out what its own scoring was doing. A record built into the tool would have shown the same thing from the start.

What governance should have looked like

Where a tool scores people for a job, the person should be able to see what was measured and how they were judged, and to challenge it. The employer should know whether the scoring treats different kinds of candidate evenly. Best practice would be for the maker and the employer to record each assessment and its basis, to give that to the candidate, and to check the scores across groups, so that if some kinds of candidate come off worse, it can be seen in the numbers. With that record, a candidate or a regulator could see what the scoring was doing, without paying for an audit to find out.

Failure Pattern: a tool ranked job candidates by reading their face and voice, on traits it may not reliably measure, and neither the candidate nor often the employer could see what was measured or how the score was reached.

Governance Principle: where a tool scores a person for a job, the person must be able to see what was measured and how they were judged and to challenge it, and the scores must be measured across groups of candidates for uneven effect.

The reference implementation of ConductRecord and MetricRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and HireVue could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →