User Handbook
Risk Classification
Classify agents per EU AI Act Art. 6 / Annex III
When is Classification Done?
Risk classification occurs in Step 3 of the agent wizard. It can be changed later in the agent drawer (triggers re-evaluation of all compliance requirements).
The Three Risk Levels
| Level | Criterion | Additional Obligations |
|---|---|---|
| Minimal | No Annex III use case | Basic governance sufficient |
| Limited | Transparency obligations (Art. 50) | Users must know they're interacting with AI |
| High | Annex III domain (Biometrics, Critical Infrastructure, Education, Employment, Public Services, Law Enforcement, Migration, Justice) | Full documentation, FRIA, mandatory responsibility chain, attestation always on-chain |
Impact of Risk Level
High-Risk means:
- Responsibility chain becomes mandatory (all 3 roles)
- FRIA must be completed before the agent can be activated
- Blockchain attestation is always on-chain (not configurable)
- Evidence packages contain additional documentation (Art. 9-15 controls matrix)
- Incident reporting obligation to supervisory authority (Art. 62)
Minimal/Limited means:
- Responsibility chain recommended but optional
- FRIA not required
- Blockchain attestation configurable
- Standard compliance sufficient