AI-Against-AI Conflict in Distributed Tactical Autonomy

Abstract

Distributed autonomy is becoming the dominant architectural paradigm for multi-agent and multi-Uncrewed Aerial Systems (UAS) that must sense, decide, and act under uncertainty, communication constraints, and partial observability. As autonomy pipelines become increasingly learned, adaptive, and coordination-driven, safe collective behavior depends on the integrity of local information and coordination interfaces through which independently acting agents compose team-level behavior. This paper argues that AI advances are not just enhancing our operations; adversaries are leveraging AI against these systems, introducing AI-against-AI conflict, in which AI-enabled, coordinated adversaries strategically shape observations, communication, and outcome feedback - through deception, information denial, communication disruption, and physical or cyber-induced perturbations - across the autonomy pipeline to induce cascading team-level coordination failures. At an autonomy-level abstraction spanning cyber and physical influence, we describe how coordinated, sequential, and stealth-constrained manipulation can compound across decision cycles and act as a force multiplier, producing abrupt team-level breakdowns without overt component failure. We conclude that future research in tactical autonomy must move beyond robustness to noise and faults, and instead develop foundations for preserving safe collective behavior under strategic, adaptive adversarial influence.