Skip to main content

Hiring Screening and Ranking

This page covers AI features used to filter, rank, or score candidates in employment workflows.

Why this is usually high-risk​

Employment decisions directly affect access to work, income, and equal treatment. If AI output influences shortlist, interview progression, or final decision, the impact is high enough that EU AI Act high-risk obligations are typically triggered.

Screen for prohibited emotion inference first (Art 5(1)(f))

Inferring the emotional state of candidates in an employment context is prohibited, not high-risk (Art 5(1)(f)) — the medical/safety exception does not cover hiring. If your feature infers emotion from voice, video, or text during screening, stop and escalate before treating it as high-risk.

Required obligations​

WHEN​

  • A feature is used for candidate filtering, ranking, or scoring.
  • A model/data/workflow change could change candidate outcomes.

MUST​

  • Treat the use case as a high-risk candidate and escalate for formal classification.
  • Block release until required high-risk controls are approved for your role (provider/deployer).
  • Check the Art 6(3) derogation: it may be lower-risk if it only performs a narrow procedural or preparatory task — but profiling of candidates makes it always high-risk. Self-classifying as not-high-risk still requires a documented assessment (Art 6(4)) and EU-database registration (Art 49(2)).
  • As deployer: inform affected workers/candidates about the use (Art 26(11)), inform workers' representatives before deployment (Art 26(7)), and support the affected person's right to explanation (Art 86).
  • Re-run classification after significant model, data, or workflow changes.

OUTPUT artifacts​

  • Classification decision record with role mapping.
  • Approval record proving required controls were completed before release.
  • Versioned change log showing reclassification triggers and outcomes.
Real-world illustration

Pre-date the AI Act and were decided under other law, but show the risk in practice:

  • iTutorGroup (US, 2023): recruiting software auto-rejected older applicants; the employer settled the EEOC's first AI hiring-discrimination suit for $365,000. (EEOC)
  • HireVue (US, 2020–21): after a complaint over biased video-interview analysis, the vendor removed facial analysis from candidate scoring — the kind of interview emotion/face inference that Art 5(1)(f) now prohibits in employment.