Strategic latency under algorithmic risk: Pakistan's divergence from the AI arms Race

Abstract

This article investigates Pakistan's evolving military artificial intelligence (AI) posture to challenge the dominant assumptions of competitive acceleration and liner diffusion in strategic and defence studies. Drawing on 14 expert interviews with experts, it provides an empirical account on how AI is interpreted and governed within a nuclearized and operationally complex security environment. The findings show that Pakistan's trajectory is not defined by technological lag or imitation, but by institutional learning derived from prior exposure to semi-autonomous systems, particularly during the War on Terror (WoT). These experiences have produced a pattern of strategic latency, in which capability development is accompanied by deliberate constraints on deployment through human oversight, threshold management, and doctrinal ambiguity. This article develops the concept of System restrain Realism (SRR) to explain how states respond not only to adversarial pressures but also to risks embedded within algorithmic systems, including misclassification and escalation under uncertainty. By demonstrating how capabilities can be decoupled from operational use, this study contributes to emerging debates on military AI by shifting attention from competition and speed to risk governance and institutional adaptation in middle-power contexts.