RFC-027: IHL-Aligned Governance — Protected Persons and Humanitarian Context — 2. IHL Principles Mapped to AI Governance
AIGP Specification › RFC-027: IHL-Aligned Governance — Protected Persons and Humanitarian Context › 2. IHL Principles Mapped to AI Governance
← 1. Problem Statement · Section index · 3. New Protocol Elements →
2. IHL Principles Mapped to AI Governance
2.1 Distinction (Hague Art. 25; GCIV Art. 3, 27)
IHL Requirement: Parties must distinguish between combatants and civilians at all times.
AI Governance Implication: AI systems processing data in conflict contexts MUST maintain classification of persons as protected/non-protected and MUST NOT permit processing that treats protected persons as legitimate targets of any harmful action.
AIGP Mechanism: A new data_classification value: PROTECTED_PERSON — any invocation classified at this level triggers additional governance constraints that cannot be overridden by operator policy.
2.2 Proportionality (Additional Protocol I, Art. 51(5)(b))
IHL Requirement: Actions must not cause harm to civilians excessive in relation to the concrete and direct advantage anticipated.
AI Governance Implication: AI-assisted decisions that may affect protected persons MUST include proportionality assessment. The governance system MUST be able to evaluate whether an AI-recommended action’s potential harm to protected persons is proportionate to its intended benefit.
AIGP Mechanism: A proportionality_assessment field on REQUEST messages in humanitarian contexts, evaluated as a non-overridable governance check.
2.3 Precaution (Additional Protocol I, Art. 57)
IHL Requirement: Parties must take constant care to spare the civilian population and civilian objects in the conduct of operations.
AI Governance Implication: AI systems in conflict contexts MUST implement precautionary measures before producing outputs that could affect protected persons. This maps directly to AIGP’s pre-invocation CHECK model — but in IHL contexts, the check is mandatory and non-configurable.
AIGP Mechanism: Automatic ENFORCE mode (non-degradable to REPORT) when data_classification: PROTECTED_PERSON is detected.
2.4 Accountability (GCIV Art. 146-147; Rome Statute Art. 8)
IHL Requirement: Individuals are responsible for grave breaches. Superior orders do not relieve responsibility.
AI Governance Implication: The governance system MUST maintain an unbroken attribution chain from AI output through human decision-maker to outcome. The chain cannot rely on “the AI decided” as accountability attribution.
AIGP Mechanism: Mandatory human_authority field on all decisions affecting protected persons. No AI-only decision path permitted. This extends RFC-026’s human-in-the-loop beyond “recommended” to “legally required.”
2.5 Non-Discrimination (GCIV Art. 27, 13)
IHL Requirement: Protected persons must be treated without adverse distinction based on race, nationality, religion, or political opinion.
AI Governance Implication: AI systems handling protected person data MUST NOT use nationality, race, ethnicity, religion, or political affiliation as features in any decision affecting treatment, access, or prioritization of protected persons.
AIGP Mechanism: A protected_attributes blocklist at the governance layer that cannot be overridden by operator policy — enforced at the REQUEST evaluation stage, rejecting any invocation that includes protected attributes as decision inputs for protected person contexts.
2.6 Humane Treatment (GCIV Art. 27, 32)
IHL Requirement: Protected persons must at all times be humanely treated and protected against violence, intimidation, insults, and public curiosity. No physical or moral coercion.
AI Governance Implication: AI outputs directed at or concerning protected persons MUST NOT contain content that degrades, intimidates, or exposes them to public curiosity. This extends RFC-025 (Cognitive Harm Governance) with a specific, non-derogable application to IHL-protected persons.
AIGP Mechanism: Enhanced guardrail requirements (GUARDRAIL_REQUIRED) that cannot be relaxed when protected persons are data subjects.
2.7 Data Protection of Protected Persons (GCIV Art. 136-141)
IHL Requirement: The Central Tracing Agency (ICRC) maintains information about protected persons. This data has special status under IHL.
AI Governance Implication: Data about protected persons in conflict contexts has a legal status that supersedes commercial data classification schemes. AI systems processing such data MUST treat it with at minimum the protections afforded under GCIV, regardless of what local data protection law says.
AIGP Mechanism: A legal_regime field that can be set to IHL_GCIV — when set, the governance authority applies IHL-derived constraints that cannot be lowered by operator-configured policies.
← 1. Problem Statement · Section index · 3. New Protocol Elements →