AI in hiring and HR: why the AI Act says high risk
Why the AI Act classes AI tools for hiring and managing employees as high risk, and what the GDPR adds on top.
On this page
In 30 seconds
- Where it happens
- CV screening, candidate ranking, scoring of test answers, allocation of tasks, or evaluation of current employees[1][2].
- What the AI Act says
- These uses appear in Annex III, so in the high-risk category, unless the derogation in Article 6 applies[1][3].
- When
- The Commission states that the rules for employment apply from 2 December 2027[4].
- The reflex
- Before adopting a tool, ask what exactly it is for, who carries out the human oversight, what data it uses, and how the people concerned are informed.
What it is
The European AI Regulation (AI Act) classes as high-risk the AI tools intended to recruit or select people, as well as those intended to make decisions about working conditions, promotion or the end of a work relationship[1].
The European Commission explains this choice. Throughout recruitment, evaluation, promotion or retention, AI systems may reproduce historical patterns of discrimination, and those that monitor performance and behaviour may undermine the rights to data protection and privacy[2]. This article explains what the texts, the Commission and the CNIL say. It does not rule on any particular case.
Straight answers
Is CV-screening software a high-risk system?
The Commission cites CV-sorting software for recruitment as an example of an AI tool for employment classed as high-risk[4]. It also gives the example of a tool that ranks and filters candidates from CVs and skills, then produces scores or rankings: recruiters remain free to review these results, but they serve as the primary basis for the decision, so the tool is classed as high-risk[2].
A derogation exists for narrow tasks. For instance, the Commission places under the Article 6 exception a system that recognises and organises CV information in a database that recruiters can search, without a significant effect on selection[2].
Why does the text also cover employees already in post?
Point 4(b) of Annex III covers systems intended to decide on working conditions, promotion or the end of a contract, to allocate tasks according to behaviour or personal traits, or to monitor and evaluate performance and behaviour[1]. The Commission gives as examples a scheduler that assigns shifts based on punctuality signals and ratings, or an online tutoring platform that automatically suspends for one month the account of a self-employed tutor whose score falls below 4 out of 5 for three consecutive weeks[2].
Are all HR tools affected?
No. The Commission places outside the scope of point 4(a) a tool that helps a candidate tailor their own CV, or a tool that supports the onboarding of new employees, which may nonetheless fall under point 4(b) if it affects working conditions, feeds into evaluations or monitors employees[2]. It also cites office booking and business travel planning as out of scope[2]. A tool that schedules interviews falls under the exception for a narrow procedural task[2].
When do the rules apply?
According to the Commission, the AI Act entered into force on 1 August 2024 and has been applicable since 2 August 2026, with exceptions. For Annex III, which includes employment, the date is postponed to 2 December 2027 by the AI Omnibus, which entered into force on 27 July 2026[4][5]. The CNIL gives the same date for Annex III[5].
Does the GDPR also apply?
Yes. According to the CNIL, the AI Regulation does not replace the GDPR, it complements it[5]. It gives automatic CV screening as an example where both texts apply[5]. The CNIL's guide reminds readers that all operations on candidates' data, from collection to deletion, are processing of personal data[6].
How it works
Article 6 sets the rule: the systems listed in Annex III are considered high-risk[3]. A derogation applies if the system does not pose a significant risk to health, safety or fundamental rights, including by not materially influencing the outcome of decision-making, and if one of these conditions is met: a narrow procedural task, improving the result of a human activity already completed, detecting decision-making patterns without replacing or influencing the human assessment already made without proper human review, or a task preparatory to an assessment[3]. The derogation never applies when the system profiles people[3]. A provider that considers its system not to be high-risk must document its assessment before placing it on the market[3].
For high-risk systems, the Commission lists requirements before market placement: risk assessment and mitigation, high-quality datasets to limit discriminatory outcomes, logging, documentation, clear information to the deployer, human oversight measures, robustness, cybersecurity and accuracy[4].
The company that uses the tool is the deployer. These obligations apply from 2 December 2027 for employment[4]. In the text published on 12 July 2024, the regulation provides several obligations for the deployer[7]:
- take measures to use the system in accordance with its instructions for use[7];
- assign human oversight to people who have the necessary competence, training, authority and support[7];
- ensure, to the extent that it controls the input data, that the data is relevant and sufficiently representative[7];
- keep the automatically generated logs under its control for at least six months, unless otherwise provided by applicable law[7].
What you do
Before adopting a tool of this kind, an HR manager or a line manager can ask these questions.
- What is the tool for: does it rank, score, filter, or only organise information? The Commission's examples show that the answer changes the classification[2].
- What do the instructions for use say, and who in the company has the competence, training and authority to carry out human oversight[7]?
- What data feeds the tool, and is it suited to its intended purpose[7]? The CNIL reminds readers that only data that is adequate, relevant and strictly necessary for selecting the candidate may be collected[6].
- How are employees and their representatives informed? Before putting into service or using a high-risk system in the workplace, the employer informs workers' representatives and the workers concerned that they will be subject to it[7].
- How are the candidates or employees whose decision the tool supports informed? The 2024 text provides for this information for deployers of Annex III systems that make or assist in making decisions about people[7].
- Is the decision fully automated? According to the CNIL, the GDPR (Article 22) gives people, in principle, the right not to be subject to a decision based solely on automated processing that significantly affects them. Exceptions exist: explicit consent, necessity for a contract, a legal provision. In these cases, the person must be informed of the logic of the decision and can ask for human intervention to contest it[8].
- Has a data protection impact assessment (DPIA) been carried out? The CNIL presumes it is required for the provider and the deployer of high-risk systems, and it can draw on the documentation required by the AI Regulation[5]. The 2024 text provides that the deployer, where applicable, uses the provider's information for this assessment[7].
- For a personality test or a serious game: the CNIL states that these tools may be used only if they have an objective, direct and necessary link with the ability to hold the job or with professional skills[6].
Pitfalls
Human oversight may be purely formal. For a background-check tool, the Commission notes that human review is part of the process but the tool's output weighs heavily on which candidates advance, which counts as material influence on recruitment[2].
Bias may come from past data. A candidate-scoring algorithm rating from 1 to 5 stars, tested as early as 2014, favoured men for technical positions, because it had learned from ten years of hiring in a male-dominated sector[9]. An employment-duration prediction model, introduced in 2015 for a call centre of 40,000 agents, tended to disadvantage residents of working-class neighbourhoods, because their longer commutes were read as a negative factor[9]. The CNIL stresses that this type of bias is harder to detect and correct[9].
Profiling is defined in Article 4 of the GDPR as processing of personal data used to analyse and predict a person's behaviour, for example their performance at work[8]. Under the AI Act, an Annex III system that profiles people never benefits from the derogation[3].
Sources
- European Commission, “AI Act Service Desk - ANNEX III”. https://ai-act-service-desk.ec.europa.eu/en/ai-act/annex-3 (accessed 14 November 2026)
- European Commission, “AI Act Service Desk - Employment”. https://ai-act-service-desk.ec.europa.eu/en/employment-0 (accessed 14 November 2026)
- European Commission, “AI Act Service Desk - Article 6: Classification rules for high-risk AI systems”. https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-6 (accessed 14 November 2026)
- European Commission, “AI Act”. https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai (accessed 14 November 2026)
- CNIL, “Entrée en vigueur du règlement européen sur l’IA : les premières questions-réponses de la CNIL”, 12 July 2024. https://www.cnil.fr/fr/entree-en-vigueur-du-reglement-europeen-sur-lia-les-premieres-questions-reponses-de-la-cnil (accessed 14 November 2026)
- CNIL, “Le guide du recrutement”, 30 January 2023. https://www.cnil.fr/fr/le-guide-du-recrutement (accessed 14 November 2026)
- Publications Office of the European Union, “L_202401689FR.000101.fmx.xml”, 12 July 2024. https://publications.europa.eu/resource/cellar/dc8116a1-3fe6-11ef-865a-01aa75ed71a1.0009.03/DOC_1 (accessed 14 November 2026)
- CNIL, “Profilage et décision entièrement automatisée”, 29 May 2018. https://www.cnil.fr/fr/profilage-et-decision-entierement-automatisee (accessed 14 November 2026)
- CNIL, “Travail et recrutement prédictif : des candidats à l’emploi inégaux face aux nouveaux usages | Linc”, 16 January 2020. https://linc.cnil.fr/travail-et-recrutement-predictif-des-candidats-lemploi-inegaux-face-aux-nouveaux-usages (accessed 14 November 2026)
Cite this article
Brain, “AI in hiring and HR: why the AI Act says high risk”, Cyberpedia, 2026. https:/
BibTeX
@misc{brain_cp532_en_2026,
author = {Brain},
title = {AI in hiring and HR: why the AI Act says high risk},
year = {2026},
howpublished = {Cyberpedia},
url = {https://startbrain.io/cyberpedia/en/articles/ai-recruitment-hr-high-risk-ai-act/},
note = {CC BY 4.0}
}