HR AI Is High-Risk: Recruitment, Screening, and the Duties That Follow

Few places put AI closer to people's lives than human resources. A model that screens applications, ranks candidates, or flags employees for performance review is making, or shaping, decisions about livelihoods. The EU AI Act recognises this, and Annex III classifies recruitment and workforce-management AI as high-risk.

The scope is broader than the obvious cases. It is not only the headline screening tool. Systems that filter applications, target job adverts, allocate tasks, monitor performance, or inform promotion and termination decisions can all fall within the employment category. The test is functional, so a feature buried inside a general HR platform can be in scope even if no one called it AI at purchase.

What makes HR distinctive is that the AI Act does not arrive into empty space. It layers on top of the GDPR and equal-treatment law, and the three interact. GDPR governs the personal data and automated-decision rights, equal-treatment law governs discrimination, and the AI Act governs the system itself. They apply cumulatively, so satisfying one does not discharge the others, and the safest analysis treats them together.

Data governance is where HR programmes struggle most. High-risk HR systems demand evidence that training and input data are relevant, representative, and examined for bias, and HR data is precisely the kind of data where historical bias hides. Meeting Article 10 here is not a formality, it is the substantive work, and it sits in tension with the GDPR's minimisation principle in ways that require documented proportionality rather than a simple answer.

The obligations stack from there: human oversight by people competent to exercise it, transparency to workers and their representatives, technical documentation, and conformity assessment and registration before deployment. For organisations running several HR AI systems across recruitment, screening, and management, prioritisation by exposure is essential.

Worked examples make the abstract concrete, from a screening portfolio at a large employer to a redundancy-selection model, where the fundamental-rights stakes are highest and the oversight design matters most.

Our whitepaper, EU AI Act Compliance in Human Resources, gives the functional scope analysis, the GDPR and equal-treatment interaction, the data-governance deep dive, and prioritisation guidance. HR AI decisions are consequential by nature, which is exactly why the Act holds them to a high standard.