RFC-028: EU AI Act Alignment — Regulatory Governance Framework — 4. High-Risk AI Systems — Mandatory Governance (Art. 6-49)
AIGP Specification › RFC-028: EU AI Act Alignment — Regulatory Governance Framework › 4. High-Risk AI Systems — Mandatory Governance (Art. 6-49)
← 3. Prohibited Practices (Art. 5) — Non-Derogable Rules · Section index · 5. Fundamental Rights Impact Assessment (Art. 27) →
4. High-Risk AI Systems — Mandatory Governance (Art. 6-49)
4.1 When risk_classification == "HIGH_RISK"
The following become mandatory AIGP protocol behaviors (non-optional):
| Article | Requirement | AIGP Mechanism | Implementation |
|---|---|---|---|
| Art. 9 | Risk management system | Governance engine + policy evaluation | risk_management_active: true on REGISTER |
| Art. 10 | Data governance | Data classification on REQUEST | data_classification field already exists — extend with EU categories |
| Art. 11 | Technical documentation | Evidence chain (RECORD + TRACE + D-DNA) | Already covered — add conformity_documentation_ref |
| Art. 12 | Record-keeping | Automatic logging (10 years retention for high-risk) | Evidence retention: retention_years: 10 |
| Art. 13 | Transparency to deployers | Disclosure metadata on VendorRecord | Add deployer_disclosure object |
| Art. 14 | Human oversight | RFC-027 human_authority pattern | Extend: human_oversight_measures field |
| Art. 15 | Accuracy, robustness, cybersecurity | Feedback loop (RFC-026) + quality monitoring | accuracy_metrics + robustness_level |
| Art. 16 | Provider obligations | Conformity declaration | ce_marking: true, eu_declaration_of_conformity_ref |
| Art. 17 | Quality management system | Evidence of QMS | quality_management_system_ref |
| Art. 19 | Automatically generated logs | TRACE spans (all 14 stages) | Already covered — mandate TRACE for HIGH_RISK |
| Art. 26 | Deployer obligations | Monitor for risks during operation | Post-market monitoring signal |
| Art. 27 | FRIA | Fundamental Rights Impact Assessment | New fria field — see §5 |
4.2 Mandatory Evidence Retention (Art. 12)
def calculate_retention(record: dict) -> int | None: regulatory = record.get("regulatory_context", {}) if regulatory.get("framework") == "EU_AI_ACT": if regulatory.get("risk_classification") == "HIGH_RISK": return 365 * 10 # 10 years — Art. 12 # ... existing logic4.3 Mandatory TRACE Mode (Art. 19)
When regulatory_context.risk_classification == "HIGH_RISK":
- TRACE is mandatory (cannot be disabled)
- Mode forced to REPORT-TRACE minimum
- All 14 governance stages must be recorded
← 3. Prohibited Practices (Art. 5) — Non-Derogable Rules · Section index · 5. Fundamental Rights Impact Assessment (Art. 27) →