AI Cyber Operations Challenge Humanitarian Law on Civilian Status

International Humanitarian Law (IHL) protects civilians from attack unless they directly participate in hostilities, a rule operationalized by a three-criterion test. This paper examines how AI-mediated civilian cyber operations specifically challenge the "direct causation" component of this test. When a civilian deploys an autonomous multi-agent cyber system, the harm produced often results from system-generated decisions made after human involvement has ceased. The existing IHL framework, which typically requires a "one causal step" standard or downstream human contributors, struggles to classify such scenarios. This creates a doctrinal tension, as the framework defaults to treating these deployments as indirect participation, potentially undermining the law's intent to capture individuals personally involved in hostilities. The paper also identifies goal-specification granularity as a key property for assessing these operations and notes that current AI governance instruments do not adequately log or report this.