PalveronPalveronDocs

Risk Classification

EU AI Act risk levels for AI systems.

The EU AI Act sorts AI systems into four risk tiers. Each tier brings a different set of obligations — registration is gated on the answer.

LevelObligations
UNACCEPTABLEProhibited — blocked at registration (Art. 5)
HIGHFRIA, transparency, human oversight, conformity assessment (Art. 6 + Annex III)
LIMITEDTransparency disclosures to end users (Art. 50)
MINIMALVoluntary code of conduct

Annex III domains (HIGH risk)

A system is HIGH-risk if it falls into one of these eight Annex III categories:

  1. Biometric identification — face recognition, behavioral analysis
  2. Critical infrastructure — energy, water, transport safety systems
  3. Education — exam scoring, admissions, learning-outcome predictions
  4. Employment — CV screening, performance evaluation, task allocation
  5. Essential services — credit scoring, insurance pricing, public-benefit eligibility
  6. Law enforcement — risk profiling, polygraph-equivalent assessment, evidence analysis
  7. Migration & border control — travel-risk scoring, document authentication
  8. Justice & democratic processes — case-law research, election interference detection

Prohibited practices (UNACCEPTABLE risk)

Registration is automatically blocked when the screening flags any of these (Art. 5):

  • Social scoring by public authorities
  • Subliminal manipulation that materially distorts behavior
  • Real-time biometric identification in publicly accessible spaces
  • Emotion recognition in workplaces and educational institutions

How Palveron uses this

  • At registration — the agent wizard asks the screening questions and maps answers to a level. UNACCEPTABLE → registration blocked. HIGH → FRIA required before approval.
  • At runtime — risk level drives default policies, mandatory blockchain attestation, and human-oversight requirements.
  • In compliance reports — Annex IV-conformant PDF exports list every agent with its tier and supporting evidence.

See the handbook page for the full step-by-step workflow inside the dashboard.

On this page