In this article8 sections
  1. Your CV-screening tool is high-risk
  2. What applies now, before December 2027
  3. What Article 26 will require from you
  4. Article 86: the right to explanation
  5. The GDPR connection you cannot ignore
  6. A practical compliance template for HR teams
  7. The date is 2 December 2027
  8. References

Your CV-screening tool is high-risk

Recruitment AI lands in the highest-scrutiny category of the EU AI Act. If you use an AI system to filter applications, shortlist, score, or rank job applicants, that system falls under Annex III, point 4(a), which lists AI used for the recruitment or selection of natural persons as high-risk. It does not matter whether the tool is a standalone product or a feature built into your ATS.

The practical consequence is that your company, as the deployer, takes on a defined set of obligations under Article 26. Since the Digital Omnibus on AI (Regulation (EU) 2026/1744), those apply to Annex III systems from 2 December 2027. Breaches of Article 26 fall in the fine tier of up to €15 million or 3% of global annual turnover under Article 99; for SMEs and start-ups, the lower of the two figures is the maximum.

The good news: Article 26 is structured and specific, you have time, and some of the groundwork is required already today.

What applies now, before December 2027

Two parts of the AI Act already apply to your recruitment AI.

AI literacy (Article 4). Since 2 February 2025, you must take measures to support the development of AI literacy of the staff who use AI systems, fitted to the tools and the context. For recruiters, that means practical training on how your screening tool works, where it fails, and when to override it. No specific level per person is required, and no certificate is needed, but keep a record of what you did.

Prohibited practices (Article 5). Since 2 February 2025, emotion recognition in the workplace is banned except for medical or safety reasons (Article 5(1)(f)). The Commission's guidelines read "workplace" as also covering candidates during selection and hiring, so video interview tools that infer candidates' emotions are prohibited. Switch such features off if your vendor offers them.

The GDPR applies in full as well, including Article 22 on automated decisions and, almost always for CV screening, a DPIA.

What Article 26 will require from you

From 2 December 2027, Article 26 sets out deployer obligations in a handful of practical areas. Each one maps to something your HR or operations team can own.

1. Use the system as intended

You must use the AI system in accordance with the provider's instructions for use. If your vendor says the tool is validated for screening software engineers and you apply it to marketing roles, you are operating outside the intended purpose. Get the provider's instructions in writing and keep them on file. Where you control the input data, make sure it is relevant and sufficiently representative for the purpose.

2. Assign human oversight

Article 26(2) requires you to assign human oversight to people who have the necessary competence, training and authority, and the support they need. In practice: a named person who understands the system's outputs, can pause or override its recommendations, and can refuse to act on a result that seems wrong. A checkbox saying "HR reviewed" is not enough. The oversight must be real, and documenting it is how you show that.

3. Monitor for unexpected behaviour

You must monitor the operation of the system on the basis of the instructions for use. If you have reason to believe its use presents a risk to health, safety or fundamental rights, you inform the provider or distributor and the market surveillance authority, and suspend use; if you identify a serious incident, you inform the provider first, then the importer or distributor and the authority (Article 26(5)). For recruitment AI, the most concrete risk is discriminatory output: the system systematically downranking women returning from parental leave, or candidates with non-Western names. You do not need a data science team. A periodic check comparing shortlist demographics against applicant pool demographics, where you lawfully can, is a credible start.

4. Keep logs

Article 26(6) requires you to keep the logs the system automatically generates, to the extent they are under your control, for at least six months. These logs are your evidence trail if a candidate challenges a decision. Check with your ATS or recruitment tool vendor: do they give you access to decision logs? If not, close that gap at the next contract renewal.

5. Inform workers and candidates

Before putting a high-risk system into use at the workplace, you must inform workers' representatives and the affected workers (Article 26(7)). And people who are subject to decisions made or assisted by an Annex III system, such as candidates, must be informed that it is used (Article 26(11)). This does not require a lengthy legal notice: a clear, plain-language statement that an AI system is used, what it does, and that a human is involved in the final decision. The application confirmation email or your careers page is a natural place for it.

Article 86: the right to explanation

Separate from Article 26, Article 86 gives a person subject to a decision taken on the basis of a high-risk AI system's output, with legal or similarly significant adverse effects, the right to clear and meaningful explanations of the role the AI played and the main elements of the decision. For recruitment, a rejected candidate can ask: what role did the AI play, and what were the main elements of the decision?

Because the right is tied to high-risk systems, plan for it by 2 December 2027 at the latest. Work with your vendor to get a plain-language description of the model's inputs and main factors, and draft a one-page internal script so your HR team can respond within a reasonable time. A response time of about a month, mirroring GDPR access requests, is a sensible benchmark.

The GDPR connection you cannot ignore

Recruitment AI processes personal data. That means the EU AI Act obligations sit on top of, not instead of, your GDPR obligations. Article 22 of the GDPR already restricts decisions based solely on automated processing that produce legal or similarly significant effects. A hiring decision is clearly significant. If your AI system is making final calls without real human involvement, you have a GDPR problem today.

The fix is the same for both: genuine human oversight, documented clearly. A private employer does not need a Fundamental Rights Impact Assessment under Article 27; that duty is limited to public bodies, private providers of public services, and deployers of credit scoring or life and health insurance pricing AI. A DPIA, though, is almost certainly required already, and it is a good place to cover the discrimination questions as well.

A practical compliance template for HR teams

Here is a straightforward four-step checklist you can start this week.

Step 1: Inventory your tools. List every piece of software your HR team uses that ranks, scores, or filters candidates. Include ATS features, LinkedIn Recruiter scoring, video interview analysis tools, and any third-party integrations. Flag which ones involve automated or AI-assisted ranking, and any that analyse emotions.

Step 2: Request vendor documentation. For each flagged tool, ask your vendor how it is preparing for 2 December 2027: its classification, its plans for conformity assessment and the EU declaration of conformity, and its instructions for use. Providers of high-risk AI systems must supply instructions for use under Article 13. A vendor with no plan is a red flag.

Step 3: Train and inform. Give recruiters practical training on the tools they use (your Article 4 measures, recorded), and add a short disclosure to your application confirmation email or careers page stating that AI-assisted screening is used and that final decisions involve human review.

Step 4: Document your oversight process. Write a one-page internal procedure: who reviews AI outputs, what criteria trigger an override, how overrides are recorded, and who runs the periodic demographic check. This document becomes the backbone of your Article 26 file.

The date is 2 December 2027

The high-risk obligations under Chapter III, Section 3 of the AI Act apply to deployers of Annex III systems from 2 December 2027. That is a fixed date. By then, your oversight procedures, candidate and worker information, and logging arrangements need to be in place. It sounds comfortable, but vendor conversations, works council consultation and staff training take longer than expected when you start from zero.

Start the inventory now. The Article 26 obligations are not technically complex. They are organisationally complex, and the Article 4 and GDPR pieces apply already.

Run your free 2-minute compliance check at comply.khairos.ai to see which AI Act obligations apply to your specific recruitment tools and get a prioritised action list tailored to your company size and sector.

References

  1. EU AI Act – Annex III: High-Risk AI Systems

  2. EU AI Act – Article 26: Obligations of Deployers

  3. EU AI Act – Article 4: AI Literacy

  4. EU AI Act – Article 5: Prohibited AI Practices

  5. EU AI Act – Article 99: Penalties

  6. EU AI Act – Article 86: Right to Explanation

  7. EU AI Act – Article 13: Transparency and Provision of Information

  8. Regulation (EU) 2026/1744 (Digital Omnibus on AI)

  9. GDPR Article 22 – Automated individual decision-making

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.

Am I compliant? Check in 2 minHow Khairos Comply works