RFC-028: EU AI Act Alignment — Regulatory Governance Framework — 11. Non-Derogable Rules — Implementation

AIGP Specification › RFC-028: EU AI Act Alignment — Regulatory Governance Framework › 11. Non-Derogable Rules — Implementation

← 10. Interaction with Existing RFCs · Section index · 12. Compliance Evidence Package →

11. Non-Derogable Rules — Implementation

Same architecture as RFC-027 ihl_rules.py:

# euai_rules.py — Non-derogable EU AI Act rules
async def evaluate_euai_rules(request: dict, context: dict) -> dict | None:
    """Evaluate EU AI Act rules. EUAI-001 through EUAI-008 are PROHIBITED.
    Cannot be overridden by operator policy."""

    if not context or context.get("framework") != "EU_AI_ACT":
        return None

    # Art. 5 prohibited practices — immediate DENY
    for rule in PROHIBITED_PRACTICES:
        if _matches(request, rule):
            return _deny(rule["id"], rule["name"], rule["description"],
                        article=rule["article"], penalty=rule["penalty"])

    # HIGH_RISK mandatory requirements
    if context.get("risk_classification") == "HIGH_RISK":
        # Art. 14: Human oversight mandatory
        if not request.get("human_oversight_measures"):
            return _deny("EUAI-HR-001", "HUMAN_OVERSIGHT_REQUIRED",
                        "High-risk AI systems require human oversight measures (Art. 14)")
        # Art. 27: FRIA required
        if not request.get("fria", {}).get("performed"):
            return _deny("EUAI-HR-002", "FRIA_REQUIRED",
                        "Fundamental Rights Impact Assessment required before deployment (Art. 27)")

    return None  # Allow to continue

---

---

← 10. Interaction with Existing RFCs · Section index · 12. Compliance Evidence Package →