Emotion recognition at work is not ‘high-risk’: it is prohibited

A comfortable reflex has taken hold in HR-techs: assuming that any use of AI in hiring is, at most, high-risk, something you classify, document and move past. There is one case where that reflex fails. Article 5(1)(f) of Regulation (EU) 2024/1689 prohibits AI systems to infer the emotions of a natural person in the areas of the workplace and education institutions, except where intended for medical or safety reasons. It is not high-risk: it is a prohibited practice.

Prohibited is not the same as high-risk

The difference is one of nature, not degree. A high-risk system under Annex III is allowed as long as you meet its obligations: documentation, risk management, human oversight, transparency. A prohibited practice under Article 5 cannot be deployed in the Union, full stop: there is no set of controls that makes it acceptable. And these are neither dead letter nor distant future, because they apply from 2 February 2025, together with the AI literacy obligation. The penalty tier is consistent with that severity: up to 35 million euros or 7% of worldwide annual turnover, the highest step in the regulation.

Where your product fits

Now bring this to an interview copilot or a video analysis. If your system infers emotions, engagement, stress or a candidate's level of "confidence" in a hiring or work context, you are not looking at something you simply classify and document: you may be looking at a directly prohibited feature. The medical or safety exception exists, but it is narrow and rarely covers a recruitment process.

Before building any further, it is worth asking three concrete questions:

• Does the system infer the person's emotional states, and not just objective data from the conversation?
• Does it do so in a hiring, assessment or work context?
• Does the medical or safety exception genuinely apply, or are you stretching it to justify the feature?