In this article10 sections
- You Are the Deployer, and That Changes Things
- What Applies to HR Today
- What Counts as High-Risk in HR?
- Your Obligations Under Article 26, From 2 December 2027
- Bias Monitoring: Good Practice You Should Start Now
- A Clarification on the FRIA: HR Is Out of Scope
- Training Records: What You Actually Need
- The Timeline
- One Action You Can Take Today
- References
You Are the Deployer, and That Changes Things
The term eu ai act for hr managers is not just a search query. It describes a legal reality. If your organisation uses AI to screen CVs, rank candidates, score employee performance, or allocate shifts, you are a deployer under the EU AI Act. Some of your obligations already apply. The heaviest ones, for high-risk HR systems under Article 26, apply from 2 December 2027: the Digital Omnibus on AI (Regulation (EU) 2026/1744) moved them from the original August 2026 date.
This is not only about your AI vendor's compliance. It is about yours.
What Applies to HR Today
Three parts of the Act already bind every HR team that uses AI.
AI literacy (Article 4, since 2 February 2025). Article 4, as amended by Regulation (EU) 2026/1744, requires providers and deployers to "take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf". It adds: "This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual." For HR, that means training and guidance that fit the tools your team uses.
The workplace emotion-recognition ban (Article 5(1)(f), since 2 February 2025). AI that infers the emotions of people in the workplace is prohibited, except for medical or safety reasons. That rules out "sentiment" or "engagement" scoring from video calls, voice or facial expressions of employees. The Commission's guidelines on prohibited practices read "workplace" to include candidates during selection and hiring, so emotion analysis of video interviews is prohibited too.
Transparency (Article 50, since 2 August 2026). If you run an AI chatbot for candidates or employees that you built yourself, it must tell people they are dealing with AI unless that is obvious. Bought chatbots are the vendor's design duty, but check that they do it.
What Counts as High-Risk in HR?
Annex III of the EU AI Act lists the categories of AI systems classified as high-risk. Annex III, point 4 covers employment, workers management, and access to self-employment:
-
AI intended to be used for the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter applications, and to evaluate candidates
-
AI intended to be used to make decisions affecting terms of work relationships, promotion or termination, to allocate tasks based on individual behaviour or personal traits, or to monitor and evaluate the performance and behaviour of persons in such relationships
If your HR stack includes any of these tools, whether it is an ATS with automated ranking, a performance management platform with predictive scoring, or a scheduling tool that allocates tasks based on individual behaviour, you are likely operating a high-risk AI system.
Your Obligations Under Article 26, From 2 December 2027
Article 26 sets out what deployers of high-risk AI must do. Here are the main obligations translated into plain HR language:
1. Use the system according to the provider's instructions (Article 26(1)). Your vendor must supply instructions for use. Read them. Keep them on file. Do not use the tool in ways it was not designed for.
2. Assign human oversight (Article 26(2)). Oversight must be given to people with the necessary competence, training and authority, and the support they need. In practice: a named person reviews AI-generated outputs before they drive a hiring or dismissal decision, and can overrule them. Document who that person is and what their review looks like.
3. Monitor and report (Article 26(5)). Monitor the system on the basis of the instructions for use. If you have reason to believe it presents a risk, inform the provider or distributor and the market surveillance authority without undue delay, and suspend use. If you identify a serious incident, inform the provider first.
4. Keep the logs (Article 26(6)). Keep the logs the system generates automatically, to the extent they are under your control, for at least six months, or longer where other law such as the GDPR requires.
5. Inform workers (Article 26(7)). Before putting a high-risk system into use at the workplace, inform workers' representatives and the affected workers. In the Netherlands, also check the consultation rights of your works council.
6. Inform the people decisions are about (Article 26(11)). People subject to decisions made or assisted by the system, such as candidates, must be informed that it is used.
Who supervises this in the Netherlands is still being settled: the Autoriteit Persoonsgegevens is the intended supervisor for AI used in work and employment, but it has no AI Act powers until the Dutch implementing law is in force.
Bias Monitoring: Good Practice You Should Start Now
The AI Act does not prescribe a bias test for deployers. Testing training data for bias is the provider's job. But monitoring the system in use is part of Article 26(5), and discrimination law and the GDPR already apply to your hiring decisions today. A simple, regular check of outcomes is the most practical way to catch problems early.
A basic workflow looks like this:
-
Export decision outputs from the AI tool (shortlists, scores, rankings) for a defined period, for example quarterly.
-
Where you lawfully hold the data, compare outcomes across groups; otherwise use the vendor's own fairness reporting.
-
Calculate pass rates, selection rates, or score distributions across groups.
-
Set an internal threshold for review. The Act sets none, so agree one with your legal adviser and apply it consistently.
-
Document findings, actions taken, and sign-off by the responsible HR lead.
Keep these records. They are your evidence if an authority, a works council or a rejected candidate asks.
A Clarification on the FRIA: HR Is Out of Scope
You may have read about the Fundamental Rights Impact Assessment (FRIA) required under Article 27. There is confusion in the market about who must complete one.
The FRIA under Article 27 applies only to deployers that are bodies governed by public law, private entities providing public services, and deployers of AI used for credit scoring of individuals or for risk assessment and pricing in life and health insurance. Private-sector HR teams are not in scope for the Article 27 FRIA obligation.
However, this does not mean you are off the hook for impact assessments entirely. The AVG/GDPR requires a Data Protection Impact Assessment (DPIA) for high-risk processing of personal data, which AI-driven HR decisions almost certainly qualify as under Article 35 of the GDPR. That obligation applies today. Run the DPIA.
Training Records: What You Actually Need
Article 4 does not require records, and no certificate is needed. But records are how you show the measures you took, and the Commission suggests keeping an internal record of trainings. For HR teams this is concrete and achievable.
Three things work well:
A training register. A simple spreadsheet listing each employee who uses an AI-assisted HR tool, the date they completed AI literacy training, and the name of the training. It does not need to be elaborate. It needs to exist and be current.
Role-appropriate training content. A recruiter using an AI screening tool needs different training than a CHRO approving headcount strategy. Cover what the tool does, its known limitations, how to identify and escalate a suspicious output, and who to contact if something goes wrong. For high-risk tools, Article 26(2) will also require that the people doing oversight are trained.
Regular refreshes. AI systems change. So do the risks. Updating training once a year, or whenever the vendor issues a significant update, is good practice.
If you are starting from scratch, ask your vendor for their instructions for use and user guidelines. That material forms the backbone of any role-specific training programme.
The Timeline
-
1 August 2024: the EU AI Act entered into force.
-
2 February 2025: prohibited practices (Article 5) and AI literacy (Article 4) apply.
-
2 August 2026: general application, transparency rules, national supervision.
-
2 December 2026: two new prohibitions apply (non-consensual sexual deepfakes and AI-generated child sexual abuse material).
-
2 December 2027: high-risk obligations for Annex III systems, including HR, apply.
For most SMEs with 20 to 200 employees, the practical steps are:
-
Now: complete an AI tool inventory, check nothing falls under Article 5, deliver AI literacy training and record it.
-
Next: run a DPIA for each high-risk HR tool, and update privacy notices.
-
In 2027: establish human oversight roles, log retention and incident channels, inform workers and your works council, and agree with vendors what documentation they will deliver.
-
2 December 2027: Article 26 duties apply.
For deployer duties under Article 26, Article 99 sets fines of up to €15 million or 3% of global annual turnover; for SMEs the lower of the two figures is the maximum. For an SME with €20 million turnover, that is a €600,000 ceiling. Article 4 has no EU fine tier; penalties for it are left to national law.
One Action You Can Take Today
Open a spreadsheet. List every AI-assisted tool your HR team uses. For each one, note: what decisions it informs, who the vendor is, whether you have their instructions for use, and whether any employee-facing notice currently exists. That list is the foundation of your AI Act work.
You do not need to solve everything today. You do need to know what you are working with.
Ready to move faster? Run the free 2-minute compliance check at comply.khairos.ai to see where your HR AI tools stand.
References
Part of our guide to High-Risk & Prohibited AI
Not sure where your company stands?
Answer 15 questions about how your team uses AI. You get a compliance score and the gaps to close first, as a PDF.