This article is the first in a broader research project examining how artificial intelligence and networked military technologies are transforming contemporary counterinsurgency and the exercise of imperial power. The project examines the changing relationship between military technology, territorial occupation, and political accountability, exploring how technological developments allow powerful states to extend their coercive capabilities beyond their borders while becoming increasingly detached from the political and material responsibilities traditionally associated with military intervention. At its centre is the concept of counterinsurgent impunity, which describes the growing capacity to exercise coercive power over foreign populations without assuming the full institutional and political burdens historically required to sustain that power.

Understanding this transformation requires examining a problem that has defined counterinsurgency long before the emergence of artificial intelligence, the production of knowledge about populations. Counterinsurgency depends on distinguishing insurgents, networks, and politically significant relationships from the civilian environments in which they operate. Historically, producing this knowledge required territorial control, sustained intelligence gathering, and the physical presence of military and administrative institutions. This article argues that artificial intelligence is changing the informational foundations of this relationship by reducing the institutional costs of making populations operationally legible, allowing military institutions to identify, classify, and act upon information at a scale and speed previously difficult to achieve. Establishing how this process works, and why it has historically depended on occupation, provides the foundation for understanding how contemporary technology is beginning to separate the exercise of coercive power from territorial presence.

I. Counterinsurgency Has Always Been a Problem of Legibility

The basic problem of counterinsurgency appears long before machine learning because the insurgent gains much of their political and military advantage by disrupting the categories through which conventional military power normally identifies an enemy. A conventional armed force organizes itself into command structures that can be recognized as military objects, whereas an insurgent movement deliberately embeds political organization and logistical support within ordinary social life, allowing combatants to disperse and relationships of political significance to appear indistinguishable from everyday mobility. The counterinsurgent can therefore possess overwhelming destructive power while still lacking the information needed to use that power selectively. The problem is not an inability to strike, but an inability to determine with sufficient confidence where the enemy ends and the civilian population begin.

Much of the history of counterinsurgency can be read as an attempt to solve this problem by manufacturing legibility. Population resettlement in Malaya did not just deprive guerrillas of food and sanctuary, it reorganized social and geographical life into forms that colonial institutions could monitor more consistently. French operations in Algeria depended on informant networks and the reconstruction of clandestine political relationships because the FLN’s political infrastructure could not be identified by destroying visible military units alone. American pacification efforts in Vietnam similarly produced an enormous bureaucracy of hamlet assessments, dossiers, local informants, and intelligence coordination, while the Phoenix Program sought to identify the individuals comprising the Viet Cong Infrastructure embedded beneath the visible armed insurgency. These campaigns differed enormously in context, doctrine,...