This is a hypothetical, illustrative example. The invented facts are: a provider in the Union sells a tool that scores job applications against a job description, ranks candidates, and recommends a shortlist to a hiring manager, who makes the final decision. On those facts, the screening in this guide finds that the tool is an AI system in scope, that the seller is a provider and the employer is a deployer, that no Article 5 prohibition is shown on the invented facts, and that the Annex III employment use case in Article 6(2) must be examined, with the Article 6(3) filter checked and the profiling condition applied. The conclusion is provisional, tied to the invented facts, and must be confirmed with a competent authority or legal adviser for any real system. This example does not determine any real recruitment tool.
What are the invented facts?
The invented facts are stated here so that the screening can be tested against them. A company, Provider A, established in a Member State, develops a tool, Tool R, that reads job applications, extracts skills and experience, scores each application against the requirements in a job description, ranks the candidates, and recommends a shortlist of ten candidates to the hiring manager. The tool uses a general purpose model as one component, with a task specific scoring layer built by Provider A. An employer, Deployer B, in the same Member State, buys Tool R, configures the job description and the scoring weights, and uses it to screen applications for a customer service role. The hiring manager reviews the shortlist, interviews the shortlisted candidates, and makes the final hiring decision. No biometric identification, emotion inference or social scoring is used. No data about protected characteristics is used as an input, on the invented facts, but the applications contain names and employment histories that can reveal characteristics. The tool is placed on the market in the Union, is put into service by Deployer B, and is used for recruitment decisions in the Union. Those facts are invented, simple and incomplete, deliberately, so that the screening steps are visible. A real recruitment tool will have more complex facts, including data sources, validation, bias testing, human oversight design, and integration with an applicant tracking system, each of which can change the screening.
How is scope and role screened?
Scope is screened first. Tool R is a machine based system that infers, from the input it receives, how to generate outputs such as scores, rankings and recommendations that can influence the hiring process, and it is designed to operate with varying levels of autonomy, with the hiring manager as the human in the loop. On the invented facts, it meets the Article 3(1) definition of an AI system, subject to the Commission guidelines on the definition being checked. It is placed on the Union market by Provider A and put into service by Deployer B in the Union, so it is within Article 2. No exclusion in Article 2 applies on the invented facts: it is not a research, military or free and open source system outside the exceptions, and it is not a system used solely for a purpose outside the Act. Role is screened next. Provider A is a provider, because it develops Tool R and places it on the market under its own name. Deployer B is a deployer, because it uses Tool R under its own authority for a professional activity. The general purpose model provider is a separate upstream provider for Chapter V purposes, and Provider A is a downstream provider for the system built on the model, with the Article 53 information and the Article 25 value chain provisions to be checked. Deployer B does not become a provider on the invented facts, because it does not substantially modify Tool R or put its name on it, but that conclusion must be rechecked if Deployer B changes the scoring weights, the data or the workflow in a way that amounts to a substantial modification.
For the broader regulatory framework, see questions and misconceptions overview.
How is the prohibition screen done?
The prohibition screen lists each Article 5 prohibition and tests it against the invented facts. Manipulative and deceptive practices: Tool R does not use subliminal techniques or manipulate behaviour, on the invented facts, so no prohibition is shown, but the marketing, the candidate information and the scoring design must be checked for deceptive presentation. Exploitation of vulnerabilities: no vulnerability exploitation is shown, but the use of employment history and personal data must be checked for vulnerability factors. Social scoring: Tool R scores applications for a specific job, not general social behaviour across contexts, so the social scoring prohibition is not shown on the invented facts, but the distinction must be recorded with reasons. Emotion inference: no emotion inference is used, so that prohibition is not shown. Biometric categorisation and identification: no biometrics are used, so those prohibitions are not shown. The new December 2026 prohibitions on non consensual intimate material and child sexual abuse material are not shown on the invented facts, because Tool R does not generate or manipulate intimate or sexual content. That screen is provisional, tied to the invented facts, and must be redone if the tool adds a new function, such as video interview analysis, emotion inference or biometric identification, which would change the prohibition analysis completely. A prohibition screen that is recorded only as not applicable, without the Article 5 list and the reasons, does not show that the screen was done.
How is the Annex route screened?
The Annex route is screened under Article 6(2) and Annex III. Annex III, point 4, covers AI systems intended to be used for recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates. Tool R, on the invented facts, analyses and filters job applications and evaluates candidates for recruitment, so it falls within that Annex III use case in outline. The Article 6(3) filter is then checked: Tool R performs more than a narrow procedural task, it does not only improve the result of a previous human activity, it does not only detect decision making patterns without replacing human assessment, and it does not only perform a preparatory task, on the invented facts, because it scores, ranks and recommends a shortlist that shapes the hiring decision. The filter also does not apply where the system profiles natural persons, and Tool R profiles candidates by scoring and ranking them, so the filter is not available on the invented facts. The provisional conclusion, tied to the invented facts, is that Tool R is a high risk system under Article 6(2), subject to confirmation. That conclusion must be checked against the Commission guidelines on high risk classification and the consolidated Annex III text, and must be confirmed with a competent authority or legal adviser. If the conclusion is confirmed, Provider A must meet the Chapter III Section 2 requirements, complete the Article 43 conformity assessment, draw up the declaration, register the system, and meet the post market duties, and Deployer B must meet the Article 26 deployer duties, including the fundamental rights impact assessment where Article 27 requires it for a public body or a private entity providing a public service, the human oversight, the monitoring and the information duties.
What open questions remain?
The open questions for Tool R, on the invented facts, include the data governance and bias testing evidence, the accuracy and robustness measures for the scoring layer and the general purpose model component, the human oversight design and the hiring manager training, the transparency to candidates and the Article 50 duties where applicable, the data protection impact assessment and the employment law position, the fundamental rights impact assessment position for Deployer B, the Article 53 information from the general purpose model provider, the post market monitoring plan, and the application date for the high risk duties, which for Annex III systems is 2 December 2027. Each open question needs an owner, a date and evidence, and the screening record should list them with the action taken. The example also leaves open whether Deployer B is a public body or a private entity providing a public service for Article 27 purposes, whether the scoring weights amount to a substantial modification if changed by Deployer B, and whether the general purpose model component has systemic risk. Those questions cannot be answered from the invented facts, and must be resolved with the real facts, the consolidated text and advice. A screening that lists no open questions, for a system as complex as a recruitment tool, should be treated as incomplete.
Continue with misconception: all AI is high risk.
What does this example not decide?
This example does not decide whether any real recruitment tool is high risk, does not determine the Article 27 position for any employer, and does not give a conformity plan for any named product. It is a hypothetical, illustrative walkthrough, built from invented facts, with a provisional conclusion tied to those facts and a confirm with authority exit. A provider or deployer with a real recruitment system should apply the screening order to its own facts, check the consolidated Regulation and the Commission guidelines, and confirm the classification and the duties with a competent authority or legal adviser. This example shows the method, not the answer for a real system.
Frequently asked questions
Is every recruitment tool high risk?
Not necessarily, but Annex III point 4 covers systems intended for recruitment and selection, including analysing and filtering applications and evaluating candidates, subject to the Article 6(3) filter. A tool that only performs a narrow procedural task, without profiling, can fall within the filter, but that must be checked against the consolidated text and guidelines. Most screening, ranking and shortlisting tools will need a careful Annex III analysis.
Does a human final decision remove the high risk classification?
No, not by itself. The Article 6(3) filter and the human oversight requirements are separate questions. A system that scores, ranks and recommends a shortlist can be high risk even where a human makes the final decision, because the system shapes the decision. Human oversight is a requirement for high risk systems, not a route out of classification.
Who must do a fundamental rights impact assessment for recruitment?
Article 27 requires a fundamental rights impact assessment for deployers that are public bodies, and for private entities providing public services, using high risk systems in defined Annex III areas, including employment. Whether a particular employer is within Article 27 depends on its status and the service, and must be checked against the consolidated Article 27 text and guidance.
When do high risk duties apply for recruitment tools?
For Annex III systems, Chapter III Sections 1 to 3 apply from 2 December 2027, under Article 113 as amended by Regulation (EU) 2026/1744. The prohibition and transparency screens, the AI literacy arrangements and the preparation of the conformity file should be done before that date, so that the system is ready when the duties apply.
Continue through this AI Act cluster
This article is an informational overview, not a legal determination. Check the current legal text and official implementation guidance for your system and role.
Consolidated Regulation (EU) 2024/1689 ↗AI Act Service Desk timeline ↗