Domain 4 of Annex III AI Act covers AI systems for recruitment and selection of workers, and for managing work-related relationships. The Commission guidelines of 19 May 2026 devote over ten pages to this domain, and the interpretation is far-reaching for HR-tech vendors and employers. Virtually any AI system that evaluates, ranks or selects candidates falls within high-risk.
For the general framework, the Article 6(3) filter and the three pitfalls, see the main article on the Commission guidelines. For the overview of all eight Annex III domains, see the hub overview.
The Two Use Cases of Domain 4
Annex III point 4 lists two use cases:
- Point 4(a): AI systems intended for recruitment or selection of natural persons, particularly for placing targeted job advertisements, analysing and filtering applications, and evaluating candidates
- Point 4(b): AI systems intended to make decisions affecting work-related relationships, for promotion or termination of employment contracts, for task allocation based on individual behaviour or personality traits, or for monitoring and evaluating performance
Together the two use cases cover the entire employee lifecycle. Anyone thinking they fall outside must be able to explain why explicitly.
Point 4(a): Recruitment and Selection
The guidelines interpret the scope of Point 4(a) broadly. It is not just about the final hiring decision, but every step in the recruitment process where AI influences the outcome.
The guidelines list four categories of activities falling within recruitment and selection: placing targeted job ads, analysing and filtering incoming applications, evaluating candidates (interviews, assessments, tests), and supporting the final decision with advice or a score.
Examples of High-Risk Systems under Point 4(a)
- AI that screens CVs and ranks candidates on fit
- AI that analyses video interviews for speech, facial expression or word choice
- AI that scores or interprets personality assessments
- AI that runs game-based assessments and advises on suitability
- AI that analyses psychometric tests and generates hire/no-hire advice
- AI that optimises job ad targeting parameters
- AI that recommends top-N candidates for a shortlist
Examples That May Fall Outside High-Risk
- An ATS (applicant tracking system) that categorises applications by vacancy and marks duplicates, without scoring or ranking candidates
- An AI system that only performs format conversion (CVs into structured fields) without evaluation
- A chatbot answering candidates' practical questions about the job without influencing selection
- Sourcing tools collecting public profiles based on objective criteria (job title, years of experience) without ranking candidates
The critical line: structural work introducing no value judgement can fall within the narrow procedural task filter. As soon as the system evaluates candidates or proposes a ranking, the filter no longer applies.
Point 4(b): Worker Management
Point 4(b) covers four categories of post-hire decisions: promotion and termination, task allocation, performance monitoring and performance evaluation.
Examples of High-Risk Systems under Point 4(b)
- AI supporting promotion or dismissal decisions with scores
- AI allocating tasks based on personality profiles or behaviour data
- AI continuously monitoring worker behaviour (productivity, communication, sentiment)
- AI automating performance reviews or generating an initial assessment
- AI in workforce management software allocating shifts or workload based on individualised profiles
- AI scoring workers on culture fit or team dynamics
Examples That May Fall Outside High-Risk
- AI suggesting roster templates without taking individual characteristics into account
- AI generating anonymous productivity statistics at team level
- AI generating meeting transcripts for archival purposes without per-employee evaluative breakdown
Sector-Specific Pitfalls
Pitfall 1: Don't Trust Vendor Claims
Many HR-tech vendors claim their system "doesn't make decisions, just supports them". The guidelines make clear this distinction is irrelevant. A system producing scores or rankings used in selection influences the outcome and falls within high-risk, regardless of whether it formally "decides".
Pitfall 2: Profiling Is Almost Always Present
Worker management and recruitment frequently fall within profiling under GDPR Article 4(4): automated processing of personal data to evaluate personal aspects. This automatically excludes the Article 6(3) filter. De facto: if you evaluate individual candidates or employees, you are high-risk.
Pitfall 3: Deployer Obligations Are Underestimated
For HR-tech, the deployer (the employer) is responsible for the FRIA (Article 27), for informing workers (Article 26 paragraph 7), for ensuring human oversight, and for proper use according to the provider's instructions for use. Many employers mistakenly assume the vendor covers all these obligations.
What to Do as an Employer
Inventory all HR-tech with AI functionality
ATS, recruitment platforms, assessment tools, performance management software, workforce management, employee monitoring. Including SaaS tools that have added AI without you knowing.
Classify per system
Does it fall under Point 4(a) or 4(b)? Is the filter relevant, or does profiling exclude it? Document the reasoning, including for filter-eligible systems.
Prepare for FRIA
For high-risk HR systems deployed as an employer in a public function or regulated sector, Article 27 FRIA applies. Start now with data, not in 2027.
Handle worker information
Article 26 paragraph 7 requires employers to inform workers and their representatives about deployment of high-risk AI. Build this into HR policy and works council procedures.
Embed AI literacy
For the broader silent revolution in HR under the AI Act, see our article on HR recruitment and the AI Act.
Test your AI against the Article 6(3) filter
Free interactive self-assessment, updated for the Commission guidelines of 19 May 2026. 9 steps, personal report with reasoning, vendor questions and next steps.