TL;DR: The EU AI Act for recruitment treats AI used for recruitment and selection as high-risk, with employer duties applying from 2 December 2027 under Regulation (EU) 2026/1744. Emotion inference in the workplace has been prohibited since 2 February 2025. The Act can reach employers outside the EU where a system's output is used there. Fines reach EUR 35 million or 7% for prohibited practices.
General information, not legal advice. Have counsel confirm how this applies to you.
Updated 28 October 2026.
Key takeaways
- Annex III, point 4 lists recruitment and selection, targeted job advertising, application filtering and candidate evaluation as high-risk uses (Annex III).
- The Omnibus, Regulation (EU) 2026/1744 of 8 July 2026 (published in the Official Journal on 24 July 2026), entered into force on the third day after publication, 27 July 2026. It moved the application of the high-risk rules for Annex III systems from 2 August 2026 to 2 December 2027, and for Annex I systems to 2 August 2028 (Official Journal; Gibson Dunn, 27 May 2026). The entry into force is also reported by Hunton.
- Article 5(1)(f) has prohibited emotion-recognition systems in the workplace since 2 February 2025, with medical and safety exceptions only (Article 5, fetched 15 October 2026). The text says "workplace" and does not mention recruitment by name; in our reading, that is the hook for interview-analysis tools, so check the scope for your own use case.
- The Commission's draft high-risk classification guidelines, published on 19 May 2026 according to Cuatrecasas, apply a "materially influences" test, so a tool can be high-risk even where a human makes the final decision. The guidelines were still in draft in mid-2026 (DLA Piper; Mayer Brown, July 2026).
- An employer that uses a high-risk system as deployer owes the Article 26 duties, including keeping logs for at least six months. An employer that rebrands or substantially modifies a system may become a provider under Article 25. Both points are covered below.
An employer that screens, ranks or evaluates candidates with software is using a system the EU AI Act classes as high-risk. The duties for that class apply from 2 December 2027, but one prohibition has applied since February 2025, and the Commission's classification guidelines are still in draft. Each of those facts changes what an HR team has to do this year.
This page sets out the classification, the dates, the deployer duties and a checklist. For the US state rules that apply to the same tools, see the dated tracker of AI hiring laws by state.
Is recruitment AI high-risk under the EU AI Act?
Definition. Under Annex III, point 4 of the EU AI Act ("Employment, workers' management and access to self-employment"), an AI system is high-risk when it is used for recruitment or selection, including targeted job advertisements, analysing and filtering applications, and evaluating candidates. The same point covers decisions on promotion and termination, task allocation and monitoring (Annex III).
The classification attaches to the use, not to the product category. The same tool is high-risk when it filters applications or evaluates candidates, and the question for an HR team is what the tool is used to do.
Classification is the first question. The prohibition is the one that already applies.
What is already banned: emotion inference at work
Article 5(1)(f) prohibits using AI systems to infer the emotions of a natural person in the areas of workplace and education institutions, except for medical or safety reasons (Regulation (EU) 2024/1689). The article's text names the workplace and does not mention recruitment explicitly; in our reading, interview analysis is the use HR teams should test against it. The prohibition has applied since 2 February 2025 (Article 5, fetched 15 October 2026).
For an employer, the practical form of that rule is a question to put to each vendor in writing. Does any feature of the tool infer emotion from a candidate's face, voice or words, and does that inference feed a score, a ranking or a proctoring flag? (For the vendor's documentation; no conclusion about any vendor.)
A tool can be lawful to screen with and still carry this prohibition for one feature. The dates below show which duties are in force now and which are not.
When do the EU AI Act hiring duties apply?
The table below sets out each rule, who it reaches and the date from which it applies. Every row carries its source. The Annex III and Annex I dates follow Regulation (EU) 2026/1744, which replaced the original date of 2 August 2026; the AI Act itself entered into force on 1 August 2024.
| Rule | Who it hits | Applies from | Source |
|---|---|---|---|
| Article 5 prohibitions, including emotion inference in the workplace | Providers and deployers of prohibited practices | 2 February 2025 | Article 5 |
| Article 4 AI literacy, softened by the Omnibus from guaranteeing literacy levels to supporting staff development | Organisations with AI-literacy duties | Obligation applied from 2 February 2025; the Omnibus (provisional political agreement 6 May 2026, confirmed by Council representatives 13 May 2026) moved it to supporting AI literacy among staff rather than guaranteeing a specific level | Gibson Dunn, 27 May 2026 |
| Article 50(2) legacy-system grace: a transparency and marking duty for generated content, not hiring-specific | Systems on the market before 2 August 2026 | 2 December 2026 | Gibson Dunn |
| Commission draft high-risk classification guidelines published | Providers and deployers classifying systems | 19 May 2026 (draft); consultation ran 19 May to 23 June 2026 | Cuatrecasas; DLA Piper |
| Regulation (EU) 2026/1744 (the Omnibus), of 8 July 2026, published 24 July 2026, in force | All operators | 27 July 2026 (third day after publication) | EUR-Lex |
| Final high-risk classification guidelines | Providers and deployers classifying systems | Expected by end of 2026 per DLA Piper (30 June 2026); in draft as of June to July 2026 | Mayer Brown, July 2026 |
| Annex III high-risk duties, including employment (Chapter III, Sections 1 to 3, for systems classified under Article 6(2) and Annex III) | Employers (deployers) and vendors (providers) of recruitment AI | 2 December 2027 | Gibson Dunn; EUR-Lex |
| Annex I high-risk duties (systems classified under Article 6(1)) | Systems covered by Annex I product legislation | 2 August 2028 | EUR-Lex; Gibson Dunn |
Cuatrecasas reports that the consultation on the draft guidelines closed on 23 June 2026. Pages published before the Omnibus agreement of May 2026 may show earlier dates, and the table above replaces them.
Dates settle when a duty starts. Whether the Act reaches a particular employer is a separate question.
Does the EU AI Act apply to a US employer?
The Act applies to providers and deployers whose system's output is used in the EU. The question an employer should ask is where the output lands, not where the employer or vendor is incorporated.
Provider or deployer? The Article 25 switch
An employer that buys a system is a deployer. An employer becomes a provider if it puts its own name or trademark on a high-risk system, makes a substantial modification to it, or changes its intended purpose. Provider duties are heavier than deployer duties.
For that reason, an employer should decide before it configures or relabels a tool whether the change could amount to a substantial modification. Once the status is settled, the deployer duties follow.
What deployers owe: Articles 26, 13 and 86, and where GDPR still applies
An employer that uses a high-risk system as deployer must use it in line with the provider's instructions, assign human oversight, ensure that the input data it controls is suitable, keep the logs the system generates for at least six months, and inform workers' representatives and affected individuals. Article 13 requires the provider to supply instructions for use, which is the document an employer should ask a vendor for. Article 86 gives an affected person a right to an explanation of a decision based on a high-risk system's output.
Private-sector employers are generally not required to carry out a Fundamental Rights Impact Assessment. A Data Protection Impact Assessment is required under GDPR.
Article 86 and GDPR Article 22 side by side
Article 86 of the AI Act and Article 22 of GDPR both address decisions made with automated processing, and an employer can be subject to both for the same hiring decision. Article 86 gives the affected person a right to an explanation of the role the AI system played. GDPR Article 22 restricts decisions based solely on automated processing.
Recorded interviews and transcripts
In our reading, GDPR transparency and lawful-basis rules apply to recorded AI interviews whatever the AI Act classification. The page on recording and GDPR basis covers recorded AI interviews in detail. Transcription tools carry their own data-protection questions, which GDPR for AI note-takers addresses.
The next section deals with the test that decides whether a tool counts as a decision-maker at all.
The "materially influences" test in the draft Commission guidelines
The Commission's draft guidelines of 19 May 2026 (Cuatrecasas) treat a system as covered where it materially influences the outcome, even if a human takes the decision. The "materially influences" reading appears in the analyses of DLA Piper, Advant NCTM and Mayer Brown (July 2026).
Under the draft, the presence of a human decision-maker does not by itself take a tool outside the high-risk class. (The draft's position, not a conclusion about any tool.)
DLA Piper (30 June 2026) expects final guidelines by the end of 2026, and Mayer Brown (July 2026) describes the guidelines as draft. This section is the first to be updated when the final text lands.
Explaining a recommendation to a candidate is a separate duty, covered in explainable AI hiring.
Which tool features touch which article?
The table below maps common interview-tool features to the articles they engage. (Vendor questions, not findings; no vendor named, no product called compliant.)
| Feature | Articles engaged | Question to ask |
|---|---|---|
| Emotion or sentiment scoring | Article 5(1)(f) | Does any score, flag or ranking depend on inferring emotion from a face, voice or words? |
| Automated scoring and ranking | Annex III point 4; Article 26 | Is the tool used to rank or evaluate candidates, and what instructions for use apply? |
| Auto-reject or auto-advance rules | Annex III; Article 14 (human oversight); Article 86 | Who can override a rule-based outcome, and can an affected person obtain an explanation of it? |
| Transcripts and logs | Articles 12 and 26 (logging); GDPR | What is logged, for how long, and who can export it? |
| Candidate-facing disclosure | Article 26(11); Article 50 where relevant | How and when is the candidate told that an AI system is used? Ask the vendor whether it treats its AI interviewer as a case for the Article 50 transparency duty. |
| Instructions for use | Article 13 | Has the provider supplied instructions for use that cover this deployment? |
Whether Article 50 reaches an AI interviewer is a question for counsel and the vendor. The table is the starting point for deciding which of an employer's tools are in scope.
Penalties under Article 99
Article 99 sets two tiers of fine. Prohibited practices carry fines of up to EUR 35 million or 7% of worldwide annual turnover. Breaches of other obligations carry fines of up to EUR 15 million or 3%. (Article 99).
Enforcement runs through authorities that each member state designates. Ask counsel which authority covers each country where you hire.
Deployer checklist and vendor questions
The checklist below is article-by-article, with an "evidence on file" column for the employer to complete.
| Step | Basis | Evidence on file |
|---|---|---|
| Inventory every AI tool used in recruitment and selection | Annex III point 4 | [ ] |
| Classify each tool by use | Annex III; draft Commission guidelines | [ ] |
| Check each tool for prohibited practices, including emotion inference | Article 5(1)(f) | [ ] |
| Obtain the provider's instructions for use | Article 13 | [ ] |
| Assign human-oversight roles | Articles 14 and 26 | [ ] |
| Set the log-retention period at six months or more | Articles 12 and 26 | [ ] |
| Notify workers' representatives and candidates | Article 26 | [ ] |
| Complete a Data Protection Impact Assessment | GDPR | [ ] |
| Review vendor contract terms, including substantial-modification terms | Article 25 | [ ] |
Eight questions to put to a vendor follow from Articles 13 and 26.
- Does the tool come with instructions for use, and do they describe this deployment? (Article 13)
- What logs does the tool generate, how long are they kept, and can the employer export them? (Article 26)
- What accuracy and bias testing has the provider documented, and does it include bias auditing by an independent party?
- How is human oversight designed, and who can override an outcome?
- Where is candidate data stored and processed?
- How are incidents reported to the employer?
- What contract terms apply if the employer configures or modifies the system, given the Article 25 switch?
- Which sub-processors handle candidate data?
Choosing between tools starts with the same classification step for each, and the AI interview platform guide describes what the category does. Buyers weighing several tools can use the buyer's guide to AI in the hiring process for the wider evaluation.
FAQ
Does the EU AI Act apply to recruitment?
Yes. Annex III point 4 classes recruitment and candidate-evaluation systems as high-risk, with duties from 2 December 2027 (Annex III; Gibson Dunn).
When does the EU AI Act apply to HR?
The prohibitions have applied since 2 February 2025. The high-risk duties for Annex III uses apply from 2 December 2027 under Regulation (EU) 2026/1744 (Official Journal).
Is AI resume screening legal under the EU AI Act?
Resume screening is not prohibited. It is high-risk, and it must meet provider and deployer obligations from 2 December 2027 (Annex III; Gibson Dunn). The prohibited part is emotion inference, not screening.
Does the EU AI Act apply to the USA?
It can apply where a system's output is used in the EU, whatever the employer's or vendor's location. Whether a particular US employer is in scope depends on where the output lands; take advice on your facts.
Is emotion detection in video interviews banned?
Emotion inference in the workplace is prohibited under Article 5(1)(f), with medical and safety exceptions, and has been since 2 February 2025 (Article 5). Whether a given feature infers emotion is a question to put to the vendor in writing.
What does Article 26 require from an employer?
An employer that uses a high-risk system as deployer must use it per the provider's instructions, assign human oversight, ensure input-data quality, keep logs for at least six months, and inform workers and candidates.
Do I need a fundamental rights impact assessment?
Generally not as a private-sector employer, though a Data Protection Impact Assessment is required under GDPR.
What are the penalties?
Fines reach up to EUR 35 million or 7% of worldwide annual turnover for prohibited practices, and up to EUR 15 million or 3% for other obligations. (Article 99).
How do EU and US state rules compare?
The EU classifies hiring AI by use and attaches duties from 2 December 2027 (Gibson Dunn). US rules sit in separate state and city laws, which the tracker linked at the top of this page sets out.
See what a decision record with reasoning and transcript looks like. See an evidence pack in a demo.
Comments
No comments yet — start the conversation.