Recruiting with AI is high-risk under the AI Act. What that actually means.

The EU AI Act classifies AI used in recruitment and selection as high-risk (Annex III). This is not an academic nuance: it comes with concrete obligations an auditor can ask for at any time.

What is required, briefly

Risk management and data governance. Transparency toward the candidate. Real human oversight on consequential decisions. And — the part most people skip — logging: you must be able to reconstruct what the system did, when, and on what basis.

Where most ATSs fail

Tools that bolted a model onto an old ATS cannot answer the simple question: "show me exactly what the AI computed for this candidate and with which model version." If that answer does not exist, the compliance is theoretical.

The practical test

Before you buy an AI ATS in the EU, ask them to show you, live: the per-inference decision log, the proof a human signs off on reject/hire, and how they generate an explanation for the candidate. If the answer is "you can do that manually," it was not built for the AI Act.

We started the other way around — from the regime to the product. The decision always stays with a human; the AI only sets the table and logs itself.