AI in recruiting and the AI Act: what is prohibited, what is high risk, and what to avoid #

AI in recruiting is useful when it supports recruiters: drafting job ads, summarizing notes, searching your own candidate database, preparing interview questions or helping with reporting. It becomes risky when it decides about people: ranking candidates automatically, filtering resumes without a meaningful human review, inferring personality or emotions, or making opaque recommendations that nobody can explain. Under the EU AI Act, some practices are prohibited and many employment-related AI systems can fall into the high-risk category. Even outside the EU, the practical rule is the same: keep a human in control, know where candidate data is processed, document what the AI does, and ask every ATS vendor whether AI features can be switched off.

Start with the data: where does it go? #

Before you discuss clever AI features, ask a simpler question: where do candidate data and prompts go. A resume, interview note or recruiter comment can contain personal data. If an ATS sends that data to an external AI service, you need to know the provider, region, retention rules, security controls and whether the data is used to train models. For sensitive HR data, "we use AI" is not enough; the vendor must be able to describe the data flow in plain language.

What is prohibited by the AI Act in recruiting? #

The AI Act prohibits several practices that are especially dangerous in employment contexts. In recruiting, avoid systems that try to infer emotions in interviews, manipulate candidates, exploit vulnerabilities, or make social scoring-style judgments. Also be extremely careful with biometric or psychological inferences. If a tool claims it can read motivation, honesty, stress or cultural fit from a video interview, treat that as a major warning sign.

Which AI features in recruiting are high risk? #

Employment is one of the areas where AI can materially affect a person's career. Tools used for screening, selection, ranking, matching, promotion or evaluation can therefore fall into a high-risk category, depending on how they are used. High risk does not always mean "forbidden", but it does mean obligations: documentation, transparency, risk management, data quality, human oversight and monitoring. If your ATS ranks candidates or recommends who should move forward, ask how the recommendation is produced and how the recruiter can challenge it.

Which uses of AI are usually lower risk? #

Lower-risk uses are typically those where AI works as an assistant and the recruiter remains the decision-maker. Examples include writing a first draft of a job ad, summarizing a candidate note, translating communication, clustering feedback from hiring managers, suggesting search keywords, or answering analytical questions over your own hiring data. Even here, the output must be checked. AI can save time, but it should not become the authority.

Ask whether AI features can be switched off #

This is the practical buying question most teams forget. If your legal, security or works-council review is not ready, can you run the ATS without AI? Can you disable only candidate-ranking features but keep harmless helpers? Are AI actions visible in the audit trail? Can users see when a text or recommendation was generated by AI? The more regulated your environment is, the more granular these controls need to be.

Related features: AI sourcing on LinkedIn · AI candidate matching · AI recommendations · Audit log

Full checklist of questions for ATS vendors →

← Back to the ATS selection guide