by Samuel White and Daniel Harker
ABSTRACT
Pre-conflict data poisoning, whereby States covertly corrupt the training data of artificial intelligence and machine learning systems before hostilities arise, exposes a profound gap in international law. Operating below recognized thresholds of armed conflict and use of force, such operations allow adversaries to quietly degrade future military capability, manipulate civilian infrastructure, and embed latent vulnerabilities across interconnected systems with minimal legal consequence. Unlike traditional sabotage, poisoned data does not destroy a single object; it contaminates entire informational ecosystems, producing cascading and potentially irreversible effects across military and civilian networks alike. Existing international humanitarian law remains tied to kinetic, event-based understandings of violence, whilst international human rights law struggles with attribution, jurisdiction, and covert cyber operations. This paper argues that the Martens Clause provides the only coherent normative restraint on such activity, supplying a humanitarian framework capable of governing emerging forms of anticipatory, non-kinetic, and technologically diffuse state competition.
