{"id":17879,"date":"2026-03-18T10:07:43","date_gmt":"2026-03-18T09:07:43","guid":{"rendered":"https:\/\/haimagazine.com\/uncategorized\/candidate-rejected-for-facial-expressions-what-did-the-system-assess-instead-of-skills\/"},"modified":"2026-03-23T13:51:38","modified_gmt":"2026-03-23T12:51:38","slug":"candidate-rejected-for-facial-expressions-what-did-the-system-assess-instead-of-skills","status":"publish","type":"post","link":"https:\/\/haimagazine.com\/en\/law-and-ethics\/candidate-rejected-for-facial-expressions-what-did-the-system-assess-instead-of-skills\/","title":{"rendered":"\ud83d\udd12 Candidate rejected for facial expressions. What did the system assess instead of skills?"},"content":{"rendered":"

the author is a member of CampusAI Community<\/em><\/p>

A candidate sits in front of the camera, answers the questions and submits the recording. A few days later, they receive a rejection. They don’t know that the decision was made before anyone at the company saw their face\u2014the system analyzed speaking rate, pauses and facial expressions, and concluded that the candidate wasn\u2019t a fit.<\/p>

This is what the hiring process looked like at Unilever, for example, from 2016 to 2020, when the company was testing automated analysis of candidates’ behavior using the HireVue tool. Many other giants used it as well, including IKEA, Hilton, Oracle, and Dow Jones.<\/p>

Tools like HireVue are developed in response to a specific problem. The decision to automate hiring rarely stems from a fascination with technology. Most often, it comes when the volume of applications starts to outstrip the time and resources available to HR.<\/p>

Unilever received over 1.8 million applications annually. Reading them all manually would be unrealistic.<\/p>

Therefore, the process was structured in stages. First, the system sorted the applications and rejected those that didn’t meet the basic criteria. Then the candidates took short cognitive tests. And finally, they recorded an asynchronous video interview, and that’s where HireVue came in.<\/p>

At this stage, the system analyzed not only the content of the answers but also on-camera behavior: pace, pauses, intonation, facial expressions. Some candidates were screened out of the hiring process before anyone at the company saw their recording.<\/p>

This is where the line is drawn. Recording alone is just a matter of organizing work. Automatic analysis of behavior\u2014facial expressions, voice, pauses\u2014is already an assessment of the candidate as a person, often based on signals that say nothing about competencies. It was precisely this element, the so-called “face reading,” that HireVue withdrew in January 2021, after a 2019 complaint by EPIC.<\/mark><\/a><\/p>

Where does the tool end and the decision begin?<\/h4>

Article 22 of the GDPR explicitly prohibits fully automated decisions that produce legal effects or otherwise significantly affect an individual. A rejection in the hiring process undoubtedly qualifies as such.<\/p>

The regulations provide three exceptions: when such a decision is necessary for entering into a contract, when permitted by law, or when the candidate gives explicit consent. However, it’s worth noting that in hiring, consent is shaky ground. Because of the power imbalance, candidates rarely feel they have a real choice, which makes such consent easy to challenge as involuntary.<\/p>

Therefore, a key “safeguard” for your company is the right to human intervention. It can’t be a mere formality. Someone on the team must actually verify the algorithm’s result, rather than just mindlessly click OK to approve a rejection list.<\/p>

The candidate must also know who is responsible for this decision and what address to contact to request a second (this time human) review of their application. When making a decision about a person, a general statement that “AI supports recruitment” is not sufficient. The candidate has the right to know exactly what was assessed and why those signals matter for the role they’re applying for. This is about a transparent process, not a right to appeal.<\/p>

And this is where the AI Act comes in<\/h4>

Behavior analysis\u2014facial expressions, tone of voice, speaking rate\u2014tries to infer more from a recording than it actually shows. Instead of measurement, there is interpretation, which easily mistakes stress for “a lack of self-confidence,” introversion for “a lack of competence,” and an accent for “poorer communication.”<\/p>

That\u2019s why the AI Act introduced a ban on the use of emotion recognition systems in the workplace and in hiring. These are systems that attempt to infer emotions or mental state based on biometric characteristics\u2014facial expressions, voice, movements. The ban took effect on February 2, 2025, and penalties for violations became enforceable on August 2, 2025\u2014up to 35 million euros or 7% of a company\u2019s global annual revenue, whichever amount is higher. This applies to every company. The same ban applies to any small company that implements a “smart” tool to read emotions from a recording at the hiring stage for an office role.<\/p>

For a company using AI in hiring, this isn\u2019t an abstract regulation from Brussels, but a very specific question: is my system evaluating a candidate for what they can do, or for how they look and how they come across on camera? If the algorithm includes analysis of facial expressions, tone of voice, or “confidence” inferred from the face, then we\u2019re no longer talking about streamlining the process, but about making decisions together with a machine about a person based on biometric signals.<\/p>

What does it mean for small businesses?<\/h4>

Unilever and a small company are two different worlds in terms of scale, but the boundary remains the same. The distinction between process automation and the automation of decisions about people applies to every company.<\/p>

A small business doesn’t get 1.8 million applications, but it has an email inbox with resumes, Excel, maybe a simple ATS (applicant tracking system) and the temptation to let “something with AI” do the candidate screening. Hiring automation in itself isn’t risky. Quite the opposite. Filtering resumes, organizing applications, or knowledge tests are ways to bring order to the chaos.<\/p>

Risk arises the moment the system starts evaluating a candidate and you stop understanding what the result is based on. An Excel test for an assistant? That’s a specific skill needed for the job. Assessing “confidence” based on facial expressions? That’s an interpretation that has nothing to do with the actual job.<\/p>

At that point, your company is no longer just automating the process\u2014it starts making decisions about a person together with an algorithm. That\u2019s the moment when responsibility doesn\u2019t disappear, it just changes form. The employer still bears it. Of course, that doesn’t mean we should forget about AI in hiring. We can still automate work safely. But automating the assessment of people… Not anymore.<\/p>

If your company uses any system that “assesses attitude” or “reads emotions” from video recordings or a candidate’s voice, it’s time to stop it and ask the vendor what exactly it measures, on what basis, and whether it already falls into the area of emotion recognition forbidden by the AI Act.<\/p>

<\/p>","protected":false},"excerpt":{"rendered":"

Can AI reject a candidate based on their facial expressions? A well-known recruitment support system did this for several years, and the world’s largest corporations used it. Today, similar practices are simply banned. That’s why every company that uses AI in recruiting should know where the line is.<\/p>\n","protected":false},"author":452,"featured_media":17828,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"rank_math_lock_modified_date":false,"footnotes":""},"categories":[805],"tags":[],"popular":[],"difficulty-level":[],"ppma_author":[843],"class_list":["post-17879","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-law-and-ethics"],"acf":[],"authors":[{"term_id":843,"user_id":452,"is_guest":0,"slug":"ewa-siciak","display_name":"Ewa Siciak","avatar_url":"https:\/\/secure.gravatar.com\/avatar\/ebee7b40b99f8697a441e9c4e843181f1ee192815264acf918eb0ef697dfbe3e?s=96&d=mm&r=g","first_name":"Ewa","last_name":"Siciak","user_url":"","job_title":"","description":"Administratywistka i entuzjastka prostych rozwi\u0105za\u0144 AI dla ma\u0142ych firm. Bada i testuje technologie, kt\u00f3re realnie odci\u0105\u017caj\u0105 przedsi\u0119biorc\u00f3w."}],"_links":{"self":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/17879","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/users\/452"}],"replies":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/comments?post=17879"}],"version-history":[{"count":1,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/17879\/revisions"}],"predecessor-version":[{"id":17880,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/17879\/revisions\/17880"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/media\/17828"}],"wp:attachment":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/media?parent=17879"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/categories?post=17879"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/tags?post=17879"},{"taxonomy":"popular","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/popular?post=17879"},{"taxonomy":"difficulty-level","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/difficulty-level?post=17879"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/ppma_author?post=17879"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}