No: since 2 February 2025, the EU AI Act bans AI systems designed to infer emotions in the workplace or in education - and a job interview explicitly falls within that scope. The ban sits in Article 5(1)(f) of Regulation (EU) 2024/1689 and applies regardless of whether the interview happens on video, on the phone, or through an AI voice interview.

This is one of the few AI Act rules classified as a "prohibited practice" - a much stricter category than the "high-risk" classification that applies to most recruiting AI systems, which mainly imposes documentation and oversight duties rather than banning the use itself.

What the AI Act actually prohibits

Under Article 5(1)(f) of the AI Act, placing on the market, putting into service, or using an AI system specifically intended to infer emotions of people in the workplace or in education institutions is prohibited, with a narrow exception for medical or safety reasons. A job interview clearly falls within the "workplace" context of the rule, since it is part of the selection process for employment.

Emotion inference under this provision means attempting to deduce an emotional state - such as nervousness, enthusiasm, uncertainty or stress - from voice, facial expression, tone, or similar signals. It does not cover a purely content-based assessment of what someone actually says, or a structured evaluation of substantive answers against pre-defined criteria.

Why the ban is so strict specifically for the workplace

The legislator justifies the special treatment of the workplace and education with the structural power imbalance involved: a candidate in a hiring process can hardly opt out of being evaluated without hurting their own chances. An AI that "detects" apparent nervousness or lack of enthusiasm and feeds that into a score would exploit exactly that dependency - on top of a technology whose reliability at detecting genuine emotion is scientifically contested.

What is still allowed

The ban targets inferring emotions, not every form of AI-assisted interview evaluation. In particular, the following remains permitted:

The distinction is not about the technology itself, but about the purpose: attempting to infer an inner emotional state triggers the ban; structured capture of what was actually said falls into a different legal category.

What is at stake for a violation

Violations of the prohibited practices under Article 5 sit in the AI Act's highest penalty tier: under Article 99 of the Regulation, fines of up to EUR 35 million or up to 7 percent of a company's total worldwide annual turnover can apply, whichever is higher. That is the same order of magnitude as the most severe GDPR fines, and clearly above the penalty range for breaches of the AI Act's high-risk obligations.

What this means for voice and chat interviews

For an AI-assisted first interview, the ban concretely means: evaluation can rely on the content of the answers - completeness, fit against pre-defined criteria, clarity of the statement - but not on an assessment of the speaker's emotional state. Vendors advertising "mood" or "personality" analysis derived from tone or vocal patterns are operating in legally risky territory that buyers should scrutinise closely before deployment.

An honest limitation applies here too: even a vendor that deliberately avoids emotion recognition cannot fully guarantee that an underlying language model never unintentionally lets tone-of-voice signals leak into its scoring. What any serious system owes its users is transparency about what the score is actually based on - and the ability to trace that back on request.

How to spot a tool that uses emotion recognition

Before deploying a voice or video interview tool, it is worth checking three things: the vendor's product description, the evaluation dimensions the system actually outputs, and the origin of the training data behind it. Warning signs include phrases like "emotional intelligence analysis", a voice-based "sentiment score", or a results report that lists categories such as "enthusiasm" or "authenticity" without a traceable link to a specific answer. A legally sound system, conversely, should be able to clearly state which content-based criteria each individual score is built on, and disclose those criteria before deployment - not only when asked after a complaint.

One practical point matters here: responsibility for making sure a deployed system complies with the ban sits with the company using the AI in its hiring process, not only with the software vendor. Buying a third-party tool is a reason to get its actual mechanics confirmed in writing, rather than relying on marketing claims alone.

Frequently asked questions about AI and emotion recognition in recruiting

Does the ban also cover facial-expression analysis in video interviews?

Yes, the ban is written to be technology-neutral and covers any AI aimed at inferring emotion, regardless of whether it analyses voice, facial expression, or both.

Can a human interviewer still pay attention to body language?

Yes. The ban targets the use of AI systems for emotion recognition, not a human's own judgment during a face-to-face conversation.

Does the medical or safety exception ever apply in recruiting?

No, practically speaking. That narrow exception is meant for things like medical diagnostic systems or safety applications, not for candidate selection.

Do employers have to actively prove their tool does not recognise emotions?

The AI Act does not explicitly impose a general proof obligation for this, but in practice it is good practice to get written confirmation of how the deployed system actually works - not least to demonstrate due diligence if a dispute arises.

How does this differ from the ban on fully automated rejections under GDPR Article 22?

These are two separate rules with different protective purposes: AI Act Article 5 bans a specific analysis method (emotion recognition), while GDPR Article 22 regulates whether a decision may be made in a fully automated way. More detail in Can AI reject candidates automatically?

Exactly when did the ban take effect?

On 2 February 2025. Unlike the AI Act's high-risk obligations, which phased in gradually through August 2026, the prohibited practices under Article 5 were among the very first rules of the Regulation to become applicable at all.

This article does not replace legal advice for a specific case. For a broader overview of AI Act obligations in recruiting beyond the prohibited practices, see EU AI Act Recruiting: Which Obligations Really Apply, and for a practical look at bias in AI interviews, see AI interview bias and fairness. For how an AI voice interview without emotion recognition is actually built, see the Atlas voice interview page; more articles live in the AI interviews topic hub. What such a conversation actually costs is broken down in What Does an AI Interview Cost in Recruiting?