The system influences who does or does not see a vacancy.
Annex III point 4a
AI in recruitment and selection under the AI Act
Route 4a concerns AI systems that influence access to work: targeted job ads, analysing or filtering applications, ranking candidates and evaluating suitability.
This page makes the recruitment route within Annex III point 4 concrete. The key question is not whether a tool contains AI, but whether the system is intended to recruit, select, filter or evaluate natural persons for work or self-employment.
Updated: May 202612 min read
When does a use case point to 4a?
Focus on decision impact. A text aid for job copy differs from a system that determines who sees a vacancy, who is ranked higher, who is excluded or who is treated as suitable.
The system analyses, filters, screens or ranks applications.
The system evaluates traits, skills, answers or suitability.
The output is used by recruiters or hiring managers in selection decisions.
Examples of recruitment AI in route 4a
The examples below are not exhaustive. They help determine which evidence items make sense per workflow.
| Use case | Why sensitive | Evidence |
|---|---|---|
| Targeted job advertisements | Determines which groups do or do not see vacancies. | Targeting criteria, channel choices, bias check, purpose limitation and candidate transparency. |
| CV parsing, matching and ranking | Influences who is considered suitable and who moves forward. | Input fields, ranking logic, match criteria, human review and override record. |
| Screening questions and knockout flows | May automatically exclude or discourage candidates before a human reviews the case. | Decision rules, exception route, human reassessment and candidate communication. |
| Assessment and interview AI | Evaluates answers, behaviour, skills or candidate suitability. | Validation, bias analysis, explainability, retention periods, oversight and objection route. |
| Recruiter assistants | May remain supporting when used for text, but can move towards 4a when used for filtering, ranking or selection advice. | Usage scenario, limitations, prompts, output review and prohibition on automatic exclusion. |
Evidence route 4a needs
For recruitment and selection, the file should show how access to work is influenced and where a human can meaningfully intervene.
Evidence item 1
Use-case register with purpose, target group, supplier, data flows and process owner.
Evidence item 2
AI Act classification note with route 4a, uncertainty points and intended use.
Evidence item 3
GDPR and bias check around applicant data, special categories, retention and data minimisation.
Evidence item 4
Human oversight instruction for recruiter and hiring manager, including override and escalation.
Evidence item 5
Candidate transparency explaining AI use, purpose, human review and rights.
Evidence item 6
Training records for recruiters, hiring managers, product teams, customer-facing teams, legal and compliance where relevant.
Borderline cases in 4a
Many recruitment tools combine text support, workflow automation and selection support. Record which function is actually used.
Generating job copy
Usually lower risk when the output is only a text proposal. Still review bias in language, inclusiveness and human approval.
Creating candidate summaries
Can be supporting, but becomes more sensitive if the summary compares candidates, ranks them or suggests suitability.
Sourcing and talent pools
Pay attention when AI determines which people become visible, which profiles are contacted or which groups remain out of view.
From recruitment use case to evidence file
Use these steps to move from scattered tool information to a demonstrable file.
Step 1
Inventory every AI function in job posting, sourcing, screening, matching, interview and assessment.
Step 2
Determine whether the function influences access to work and record route 4a or an outside-scope rationale.
Step 3
Connect each use case to GDPR lawful basis, data categories, retention and bias risks.
Step 4
Describe where recruiters and hiring managers review, deviate and record deviations.
Step 5
Align candidate communication and role-based AI literacy with the actual workflow.
Annex III point 4a
Use route 4a as a working layer inside the HR-AI hub
Start with the whitepaper for context, then use the Evidence Pack to record matching, screening and selection as concrete use cases.
Sources and related routes
Use this route together with the broad hub, the classification check and the template pack.
Annex III point 4 hub
Use the broad hub to assess route 4a and 4b together.
Open routeAnnex III point 4 classification check
Get an initial indication whether your use case points to 4a, 4b or outside scope.
Open routeHR-AI Evidence & Readiness
Read the whitepaper as the explanation layer above the template pack.
Open routeHR-AI Evidence Pack
Download editable Word and Excel templates for route 4a and 4b.
Open routeAI in worker management, monitoring and performance evaluation
Compare this route with the other Annex III point 4 route.
Open routeNext route
Next steps outside Responsible AI Platform
Responsible AI Platform remains the source layer. For implementation or training, connect this route to the right environment.
Embed AI
Embed AI for recruitment-AI evidence
For classification, gap analysis and evidence around matching, screening, job ad targeting and candidate evaluation.
LearnWize
LearnWize for role-based training
For AI literacy and training records for recruiters, hiring managers and HR teams.
FAQ about route 4a
Does every ATS feature fall under Annex III point 4a?
No. An administrative ATS feature differs from AI that analyses, filters, ranks or evaluates applications. Look at the concrete function and intended use.
Is CV matching always high-risk?
When CV matching is used to rank, filter or evaluate candidate suitability for access to work, route 4a is likely relevant. Record the exact decision impact.
What should human oversight in selection AI mean?
The recruiter or hiring manager must be able to understand the output, review it, set it aside and record deviations. Merely accepting an automated ranking is not meaningful oversight.
What communication to candidates is needed?
Explain that AI is used, for what purpose, what human review exists and where candidates can go with questions or reassessment requests. Align this with the real workflow.
Sources
This page uses the AI Act, the draft high-risk classification guidelines and relevant supervisory context as its basis.