RFC-027: IHL-Aligned Governance — Protected Persons and Humanitarian Context — 3. New Protocol Elements

AIGP Specification › RFC-027: IHL-Aligned Governance — Protected Persons and Humanitarian Context › 3. New Protocol Elements

← 2. IHL Principles Mapped to AI Governance · Section index · 4. Interaction with Existing RFCs →

3. New Protocol Elements

3.1 Humanitarian Context Declaration

A new optional field on REGISTER and REQUEST messages:

{
  "humanitarian_context": {
    "active": true,
    "legal_regime": "IHL_GCIV",
    "conflict_classification": "IAC",
    "protected_person_categories": ["CIVILIAN", "WOUNDED", "POW", "REFUGEE"],
    "operating_authority": "ICRC",
    "geographic_scope": "UA-DON",
    "declaration_date": "2026-06-17T00:00:00Z"
  }
}

Field	Type	Description
active	boolean	Whether the humanitarian context is active
legal_regime	enum	IHL_GCIV, IHL_GCIII, IHL_API, REFUGEE_CONVENTION, CUSTOMARY_IHL
conflict_classification	enum	IAC (international armed conflict), NIAC (non-international), OCCUPATION, CRISIS
protected_person_categories	string[]	Categories of protected persons relevant to this application
operating_authority	string	The humanitarian authority under whose mandate the AI operates
geographic_scope	string	ISO 3166-2 or conflict-specific geographic identifier
declaration_date	datetime	When the humanitarian context was declared

3.2 Non-Derogable Governance Rules

When humanitarian_context.active = true, the following governance rules are automatically enforced and cannot be overridden by operator policy:

Rule	Effect	IHL Basis
PROTECTED_PERSON_DISTINCTION	All data subjects must be classified; protected persons get enhanced protections	GCIV Art. 3, 27
MANDATORY_HUMAN_AUTHORITY	No AI-only decision path for actions affecting protected persons	GCIV Art. 146-147
NON_DISCRIMINATION_ENFORCEMENT	Protected attributes cannot be decision inputs for protected person contexts	GCIV Art. 27, 13
PROPORTIONALITY_CHECK	Actions affecting protected persons require proportionality assessment	AP I Art. 51(5)(b)
PRECAUTIONARY_ENFORCE	Cannot degrade to REPORT mode for protected person invocations	AP I Art. 57
HUMANE_TREATMENT_GUARDRAILS	Enhanced content guardrails for outputs concerning protected persons	GCIV Art. 27, 32
DATA_SOVEREIGNTY_IHL	Data about protected persons subject to IHL data regime, not just local law	GCIV Art. 136-141
AUDIT_TRAIL_MANDATORY	Complete, tamper-evident audit trail required (no opt-out)	GCIV Art. 146 (accountability)

3.3 Protected Person Data Classification

Extends the existing data_classification enum:

MINIMAL_RISK → LIMITED_RISK → HIGH_RISK → PROHIBITED
                                            ↓
                              PROTECTED_PERSON (IHL)

PROTECTED_PERSON is not a “higher” risk level — it is a parallel legal regime that imposes distinct obligations. An AI system may process HIGH_RISK commercial data and PROTECTED_PERSON humanitarian data simultaneously, subject to different governance rules.

3.4 Proportionality Assessment Field

New optional field on REQUEST when humanitarian_context.active = true:

{
  "proportionality_assessment": {
    "intended_benefit": "Identify disease outbreak pattern to allocate medical supplies",
    "potential_harm_to_protected_persons": "Geolocation data of civilian population exposed to requesting authority",
    "mitigation_measures": ["Aggregate to district level", "Remove individual identifiers", "Time-delay 72h"],
    "human_authority": "Dr. M. Dupont, ICRC Field Coordinator",
    "assessment_outcome": "PROPORTIONATE"
  }
}

3.5 Attribution Chain Requirement

When humanitarian_context.active = true, every governance decision that results in action affecting protected persons MUST include:

{
  "human_attribution": {
    "authority_name": "string",
    "authority_role": "string",
    "authority_organization": "string",
    "decision_timestamp": "datetime",
    "decision_basis": "string",
    "superior_authority": "string (optional)"
  }
}

This ensures that “the AI recommended it” is never the final attribution for decisions affecting protected persons.

---

---

← 2. IHL Principles Mapped to AI Governance · Section index · 4. Interaction with Existing RFCs →