By Lee Flanagan
✨ AI Summary:
- EU high-risk AI rules (effective August 2, 2026) mandate human oversight in all recruitment AI decisions, including CV screening, ranking, and automated interviews—no system can operate unsupervised.
- Candidates must be informed before AI is used, told what role it plays, and given the right to request human review of outcomes, making transparency and explainability non-negotiable compliance requirements.
- AI cannot infer personality, emotion, or stress from voice tone, facial expression, or body language; this prohibition addresses real bias risks where cultural or nervous variation gets misclassified as job-relevant data.
- Only one in four companies have begun compliance preparation despite 39% already using AI in HR, creating urgent legal and fairness exposure for employers worldwide using EU-based candidate data.
A jobseeker in Vienna sat down for an AI-run interview and got cut off mid-sentence. The system moved on before he finished an answer, treated a short pause as the end of a response, and mangled its own transcript of what he said. When he tried to ask the interviewer a question, it ignored him and repeated text from the company’s careers page instead. EURES, the EU’s job-mobility information service, uses his account, in which he said the process left him feeling “dehumanized,” to illustrate exactly what the EU’s new AI recruitment rules exist to stop.
The EU classifies AI used in recruitment and other employee-related decisions as high-risk. EURES guidance, reported by Informat.ro, says that from August 2, 2026, companies running these systems must carry out periodic checks for risk and discrimination, keep documentation, and guarantee a person can intervene in an AI-influenced hiring decision. Read the requirements closely and they do not describe a new bar. They describe the standard a rigorous interview process should have cleared all along: tell candidates what is happening, keep a human accountable for the outcome, and never let a machine turn tone of voice into a personality score.
What August 2, 2026 Actually Requires
EURES lays out the mechanics plainly. Employers using AI to filter CVs, rank candidates, or run automated interviews must monitor those systems throughout their use, not just at rollout, and keep technical records showing the checks happened. Candidates must be told beforehand that AI is involved, what role it plays in the decision, and that they can ask for a person to review the outcome. EURES is explicit that the automated tool cannot serve as a full substitute for human responsibility in a decision over whether someone gets or loses a job. Candidates can also find out what data was collected and, in many cases, have it deleted once the process ends.
None of this denies that AI screening tools deliver on efficiency. EURES itself notes that filtering CVs, ranking candidates, and running automated interviews can speed up selection and cut costs. The EU’s rule does not dispute that. It disputes the idea that speed and cost are the only things worth measuring when the same tool decides who gets an interview.
The Vienna Interview Is What Failed Oversight Looks Like
Every element of that Vienna interview maps onto a rule the EU has just written down. A system deciding on its own that a pause meant an answer was over, without anyone able to override it, is exactly what the human oversight requirement exists to prevent. A candidate unable to get a real answer to his own question is exactly the transparency failure EURES has in mind when it insists a person answer for the outcome. The rule does not ban automated interviews. It bans running them unsupervised.
EURES is equally specific about what AI cannot do inside an interview. It cannot read facial expression, voice tone, or body language to infer emotion, stress, or personality, and then turn that inference into a verdict. That prohibition addresses a real technical problem, not a hypothetical one. A system that reads a long pause as low confidence, or a flat tone as disinterest, is guessing at psychology from signals that vary by person, culture, and simple nerves, then presenting the guess as data.
Only About One in Four Companies Has Started Preparing
The adoption curve is already ahead of the compliance curve. EURES cites an estimate that 39% of companies were using AI in HR activities by the end of 2025.
Only about one in four companies have begun preparing for the requirements now in force. That shortfall leaves a large population of employers running high-risk systems without the checks, documentation, or human-review processes the law now demands of them.
The obligations do not stop at the EU’s border either. EURES specifies that a company established outside the EU falls under the requirements when its AI system is used on people inside the Union or affects them. An employer or technology provider based in the United States, Asia, or elsewhere cannot opt out because its head office sits outside the bloc. If your recruitment AI ever touches a candidate physically located in the EU, the classification applies to you too.
The Compliance Question Is an Explainability Question
Strip away the legal language and the requirements are simple. A candidate should know when AI is involved, a person should be able to explain and revisit the outcome, and no one should rate someone’s character from a shake in their voice. None of that is a technical achievement. It is discipline good interviewers already practice, whether or not a regulator asked them to.
Our read is that the EU has not created a new standard for interviewing. It has written into law the difference between an interview that produces evidence a person can defend and one that produces a number nobody can explain. Interviewers who can already put a rejection into plain, defensible language will meet these requirements without changing much. Those whose process leaned on AI to make the call, and to make it sound scientific, now face a compliance problem as well as a fairness one.
A candidate who cannot get a straight answer from an interview bot has not experienced automation done well. He has experienced a company that let software make a decision no one in the building can explain. The EU has just turned that experience into legal exposure for the company that ran it.
Original reporting: Informat.ro.
Frequently asked questions
Does the EU’s high-risk classification cover only automated interviews, or CV screening too?
EURES classifies AI used for recruitment, performance evaluation, and other employee-related decisions as high-risk. That covers CV filtering and candidate ranking as well as automated interviews, not interview tools alone.
Can a candidate refuse an AI interview outright, or only request a human review afterward?
EURES sets out a right to be informed that AI is being used and a right to request that a person review the resulting decision. It does not describe a separate right to refuse the AI step itself.
If a company is headquartered outside the EU, does using an AI screening tool on EU applicants still trigger these obligations?
Yes. EURES specifies that a company established outside the EU falls under the requirements once its AI system is used on individuals inside the Union or produces effects on them.
What happens to a candidate’s interview data once the recruitment process ends?
EURES states candidates have the right to find out what data was processed about them and, in many cases, can request that it be deleted once the recruitment process concludes.