---
title: "The AI Act and recruitment: what changes for an agency or an HR team"
description: "The EU AI Act classifies tools that screen or evaluate candidates as high-risk. The timeline after the digital omnibus (2 December 2027), deployer and provider obligations, practices already banned: what a recruitment agency or an HR team needs to know."
url: https://marvins.ai/en/blog/ai-act-recruitment
language: en
published: 2026-11-11
author: "Benjamin Piloquet"
---
# The AI Act and recruitment: what changes for an agency or an HR team
The EU AI Act classifies tools that screen or evaluate candidates as high-risk. The timeline after the digital omnibus (2 December 2027), deployer and provider obligations, practices already banned: what a recruitment agency or an HR team needs to know.
The **EU AI Act** classifies as **high-risk** the AI systems used to recruit: targeted job ads, application screening, candidate evaluation. For the agency or HR team that uses them, the obligations apply on **2 December 2027**: the digital omnibus, in force since 27 July 2026, moved the date from 2 August 2026.
This article explains what the regulation counts as high-risk, what falls on the software vendor and what falls on the user, what already applies, and what remains to prepare.
_Timeline checked on 11 November 2026, against Regulation (EU) 2026/1744 of 8 July 2026 (digital omnibus on AI) and the European Commission’s AI Act page, updated on 20 October 2026. This article is not legal advice._
## Key takeaways
- **Recruitment is a high-risk use.** Annex III of the regulation covers AI systems used to recruit or select people, and those used to decide on promotion or the end of a work relationship.
- **The date to remember is 2 December 2027.** That is when high-risk obligations apply to these uses, after the delay adopted in 2026.
- **Some rules already apply.** Recognising a candidate’s emotions has been prohibited since 2 February 2025, and a chatbot must say it is an AI since 2 August 2026.
- **The agency is a “deployer”.** The vendor is responsible for the system’s compliance; the agency is responsible for how it uses it: human oversight, data, logs, informing candidates.
- **The GDPR still applies.** The AI Act adds to the rules on automated decisions and informing candidates; it does not replace them.
## The AI Act timeline for recruitment
Regulation (EU) 2024/1689 entered into force on 1 August 2024, but its rules apply in stages. For recruitment, three dates matter: 2025 for the bans, 2026 for transparency, and December 2027 for the high-risk obligations, after the delay decided by the digital omnibus.
| Date | What applies |
| --------------- | ---------------------------------------------------------------------------- |
| 1 August 2024 | Entry into force of the regulation |
| 2 February 2025 | Prohibited practices (including emotion recognition at work) and AI literacy |
| 2 August 2025 | Rules for general-purpose AI models and governance |
| 2 August 2026 | Transparency (Article 50): telling people they are interacting with an AI |
| 2 December 2027 | High-risk systems in Annex III, including recruitment |
| 2 August 2028 | High-risk systems built into regulated products (Annex I) |
The text is on EUR-Lex (Regulation (EU) 2024/1689).
### What the digital omnibus changed
On 19 November 2025, the Commission proposed to simplify the implementation of the AI Act. The text was adopted as Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July 2026, and entered into force on 27 July 2026.
For recruitment, its main effect is the delay: high-risk obligations for Annex III move from 2 August 2026 to **2 December 2027**. The bans already in force do not change. The omnibus also lightened the registration of systems and reworded Article 4: providers and deployers take measures to develop the AI literacy of the people who use these systems on their behalf.
## What Annex III counts as high-risk
Point 4 of Annex III covers employment. It classifies as high-risk the AI systems intended for recruiting or selecting people, in particular to place targeted job ads, analyse and filter applications, and evaluate candidates. It also covers promotion, termination of a contract and performance evaluation.
In practice, an AI feature that scores a candidate against a job, ranks applications or filters them falls within that scope. A candidate database, a keyword search or an interview calendar, without AI that evaluates, does not.
### The Article 6 exception, and its limit
Article 6 provides a way out: an Annex III system is not high-risk if it does not materially influence the decision, for instance if it performs a narrow procedural task or a preparatory task. A provider relying on this must document its assessment before placing the system on the market.
The limit is written in the same article: an Annex III system that performs **profiling** of people is always high-risk. Evaluating a candidate’s professional abilities is, as a rule, profiling within the meaning of the GDPR. A candidate score is therefore unlikely to escape the high-risk category.
## Provider or deployer: who does what
The regulation splits obligations between two roles. The provider develops the AI system and places it on the market under its own name: the software vendor. The deployer uses it under its own authority: the recruitment agency or the hiring company. Most technical obligations fall on the first, how the tool is used falls on the second.
**The provider** designs a system that allows human oversight and keeps logs, documents it, has its conformity assessed, registers it in the EU database and gives the deployer instructions for use.
**The deployer**, under Article 26:
1. uses the system **according to the provider’s instructions**;
2. gives **human oversight** to people with the necessary competence, training and authority;
3. makes sure the **input data** is relevant and representative, where it controls that data;
4. **monitors how the system works**, informs the provider and suspends use if it presents a risk;
5. keeps the **logs** generated by the system for at least **six months**;
6. **informs workers’ representatives** and the workers concerned before putting the system into service at the workplace, when it is an employer;
7. **informs the people** subject to a decision taken or assisted by the system.
Article 86 adds a right for candidates: to obtain from the deployer a clear explanation of the AI’s role in a decision that affects them.
### When an agency becomes a provider
Article 25 turns a deployer into a provider in three cases: it puts its name or trademark on a high-risk system, it substantially modifies one, or it changes the intended purpose of an AI system that then becomes high-risk. Using a general-purpose AI assistant to rank applications may fall into the last case: check with your adviser before making it a method.
## What already applies in recruitment
Several AI Act rules already apply to recruiters. Since 2 February 2025, Article 5 has prohibited emotion recognition at the workplace, and Article 4 has expected teams to understand the AI they use. Since 2 August 2026, Article 50 has required that people know when they are interacting with an AI.
**Emotion recognition is prohibited from recruitment onwards.** The Commission’s guidelines on prohibited practices say it plainly: the notion of workplace covers candidates during selection, and using emotion recognition during recruitment is prohibited. A video interview tool that infers a candidate’s stress or enthusiasm from their face or voice falls under this ban.
**A chatbot says it is an AI.** An agent that answers candidates on a careers site or by message must be designed so that the person knows they are interacting with an AI.
## The penalties
Article 99 sets maximum fines by category of breach. The highest covers prohibited practices: up to €35 million or 7% of worldwide annual turnover. Breaches of deployer obligations under Article 26 go up to €15 million or 3%. For an SME, the cap is the lower of the two.
Penalties are imposed by the national authorities each Member State designates. They add to those of the GDPR, which still apply to any candidate data.
## The AI Act and the GDPR: both apply
The AI Act does not replace any GDPR rule. An AI screening tool remains subject to Article 22, which in principle prohibits rejecting a candidate through a fully automated decision, and to the duty to tell every candidate what you do with their data, automated screening included.
The detail is in our article [GDPR in recruitment: how long to keep a CV, and what to tell candidates](https://marvins.ai/en/blog/gdpr-recruitment-cv-retention).
## What an agency can prepare before December 2027
The delay leaves a little over a year. The work comes down to five steps, to take with your DPO or adviser, and it starts with knowing which AI tools your team really uses, including the ones it adopted without saying so.
1. **List the uses of AI** in recruitment: screening, scoring, writing, interviews, chatbots, and the tool behind each.
2. **Classify each use**: prohibited, high-risk, transparency only, or out of scope.
3. **Ask each vendor** for its instructions for use, its classification of the system and the list of its AI providers.
4. **Name who exercises human oversight** and train them: who reviews the applications a tool puts at the bottom of the list, who can suspend the tool.
5. **Update the information given to candidates** and, if you are an employer, plan to inform workers’ representatives.
To compare tools on these points, see our [checklist for compliant recruitment software](https://marvins.ai/en/blog/gdpr-ai-act-compliant-recruitment-software).
## In Marvin
Marvin helps you prepare these steps, by documenting its AI features and leaving the decision to your team. Classifying your uses and your organisation’s compliance remain your responsibility, with your DPO or adviser.
- **The AI system sheet, on request.** The AI providers and the AI system sheet are part of the Trust Pack, with the DPA, the security measures and the continuity plan. The providers are also named on the [responsible AI](https://marvins.ai/en/responsible-ai) page.
- **No training on your data.** Marvin does not use your data to train AI models; its providers work on API offers that exclude training by default.
- **Sending approved by you, by default.** When the AI assistant prepares a message or a sequence enrolment, you see the preview and approve it before Marvin runs it.
- **A score rejects no one.** An automation based on a score can move a candidate to an in-progress stage above a threshold you set; below that threshold, nothing happens. It can neither reject them nor mark them as placed.
- **The history of each profile** keeps the team’s actions, with their author and time.
Data is hosted in France, with Scaleway in Paris; sub-processors, AI providers included, are listed in the DPA. See the [security and Trust Pack](https://marvins.ai/en/security) page.
## FAQ
**When does the AI Act apply to recruitment tools?**
The obligations for high-risk systems in Annex III, which includes recruitment, apply from 2 December 2027. The digital omnibus on AI, in force since 27 July 2026, moved this date from 2 August 2026. Prohibited practices have applied since 2 February 2025.
**Is recruitment software a high-risk AI system?**
Not as such. Annex III covers AI systems intended for recruiting or selecting people, in particular to place targeted job ads, analyse and filter applications, or evaluate candidates. An AI feature that ranks or scores candidates falls within that scope; a candidate database without AI that evaluates does not.
**Who is the deployer and who is the provider?**
The provider develops the AI system and places it on the market under its own name: the software vendor. The deployer uses it under its own authority in its business: the agency or company that recruits. A deployer becomes a provider if it puts its name on the system, substantially modifies it or changes its intended purpose.
**What must an agency using a high-risk AI tool do?**
Use it according to its instructions, give human oversight to competent people, make sure the data it feeds in is relevant, monitor how it works, keep its logs for at least six months, inform the candidates concerned and, as an employer, inform workers’ representatives before putting it into service at the workplace.
**What is already banned in recruitment?**
Since 2 February 2025, Article 5 has prohibited AI systems that infer the emotions of a person at the workplace, except for medical or safety reasons. The Commission’s guidelines state that this applies from the recruitment process onwards: analysing a candidate’s emotions in a video interview is prohibited.
**What penalties does the AI Act set?**
Up to €35 million or 7% of worldwide annual turnover for a prohibited practice, and up to €15 million or 3% for a breach of deployer obligations. For an SME, the cap is the lower of the two amounts.
## FAQ
### When does the AI Act apply to recruitment tools?
The obligations for high-risk systems in Annex III, which includes recruitment, apply from 2 December 2027. The digital omnibus on AI, in force since 27 July 2026, moved this date from 2 August 2026. Prohibited practices have applied since 2 February 2025.
### Is recruitment software a high-risk AI system?
Not as such. Annex III covers AI systems intended for recruiting or selecting people, in particular to place targeted job ads, analyse and filter applications, or evaluate candidates. An AI feature that ranks or scores candidates falls within that scope; a candidate database without AI that evaluates does not.
### Who is the deployer and who is the provider?
The provider develops the AI system and places it on the market under its own name: the software vendor. The deployer uses it under its own authority in its business: the agency or company that recruits. A deployer becomes a provider if it puts its name on the system, substantially modifies it or changes its intended purpose.
### What must an agency using a high-risk AI tool do?
Use it according to its instructions, give human oversight to competent people, make sure the data it feeds in is relevant, monitor how it works, keep its logs for at least six months, inform the candidates concerned and, as an employer, inform workers’ representatives before putting it into service at the workplace.
### What is already banned in recruitment?
Since 2 February 2025, Article 5 has prohibited AI systems that infer the emotions of a person at the workplace, except for medical or safety reasons. The Commission’s guidelines state that this applies from the recruitment process onwards: analysing a candidate’s emotions in a video interview is prohibited.
### What penalties does the AI Act set?
Up to €35 million or 7% of worldwide annual turnover for a prohibited practice, and up to €15 million or 3% for a breach of deployer obligations. For an SME, the cap is the lower of the two amounts.