AI Weapons Could Shrink Decision Time and Unravel Strategic Stability, Study Warns

Artificial intelligence is quietly rewriting the arithmetic of nuclear-age deterrence, and a new open-access review argues that the world’s major powers are drifting toward a destabilizing dynamic that scholars are only beginning to name. Writing in Discover Global Society, researcher Chick Edmond of Old Dominion University examines whether and how AI-enabled autonomous weapons could erode strategic stability among the United States, China, and Russia. The study’s central claim is deliberately narrow: while no one has yet demonstrated that autonomous systems have caused a major escalation, the doctrines and operational trends of all three great powers are converging on capabilities consistent with what Edmond calls an AI security dilemma.

The classical security dilemma, articulated by Robert Jervis in 1978, holds that states seeking to improve their own security often make others feel less safe, triggering spirals of mistrust even when everyone acts defensively. Edmond extends this logic to machine intelligence. Because AI systems are opaque, fast, and dual-use, a defensive algorithm looks identical to an offensive one from the outside. A weapon trained to recognize enemy tanks can be repurposed for civilian targets, and the same navigation software guides defensive and offensive platforms alike. States therefore assume the worst about each other’s AI programs and accelerate their own, a race that no rational actor can easily opt out of.

The study identifies three causal mechanisms through which autonomy could destabilize crises. The first is the compression of decision-making time. Machines process information and act in fractions of a second, while human deliberation takes minutes, hours, or days. In high-intensity conflict, commanders may be forced to pre-delegate authority to algorithms simply to keep pace, and once that authority is granted, autonomous systems can make escalatory choices that adversaries cannot distinguish from deliberate aggression. Edmond illustrates the stakes with the 1983 Stanislav Petrov incident, in which a Soviet officer correctly judged a false missile alert to be a malfunction. Under machine-speed timelines, he argues, there may be no window for a similar human override.

The second mechanism is the fragmentation of escalation pathways. Autonomous systems proliferate across air, sea, cyber, and land domains, each capable of small, localized actions. Unlike Herman Kahn’s tidy escalation ladder of discrete rungs, this produces what the study calls cumulative or salami-slicing escalation: many individually minor actions that together erode an adversary’s position until a dramatic response becomes the only option. Because these actions are decentralized, leaders struggle to determine whether an incident reflects intentional coercion, an operational accident, or emergent behavior produced by the algorithm itself, and worst-case thinking fills the interpretive vacuum.

The third mechanism is the erosion of meaningful human control. Autonomy exists on a continuum, from human-in-the-loop systems where operators approve every engagement, to human-on-the-loop systems that act unless overridden, to hypothetical human-out-of-the-loop weapons that select and strike targets on their own. No fully autonomous weapon is known to be operational, but the study warns that technical veto authority becomes hollow when a human has seconds to exercise it. Deterrence depends on credible signaling and attribution, and if a state cannot reliably assign responsibility for actions taken by its own machines, it loses the ability to communicate intentions at all.

Applying a structured coding framework to official doctrine documents, national AI strategies, and policy papers published between 2016 and 2024, the study finds distinct national flavors of the same underlying trend. The United States pursues decision superiority, the goal of making faster and better decisions than any adversary, while formally committing to appropriate levels of human judgment. Edmond describes this as a control paradox: the very speed and complexity that deliver decision superiority narrow the window for human intervention, leaving each commander to define what appropriate judgment means in real time. The Department of Defense’s Chief Digital and Artificial Intelligence Officer oversees development, but critics note that governance structures have not kept pace with fielded systems.

China frames autonomy within intelligentized warfare, a concept its defense white papers describe as the main type of future war. People’s Liberation Army theorists envision AI woven through surveillance, command and control, cyber operations, and unmanned platforms, with semi-autonomous systems enabling persistent, low-intensity coercion below the threshold of armed conflict. Yet the study stresses that extreme opacity surrounds how these doctrinal commitments translate into operational capability. Russia, constrained by economics and demographics, embraces automation to offset conventional weakness and pairs it with a doctrine of risk acceptance, explicitly designating nuclear weapons as tools of escalation management while delegating authority to local commanders in hybrid operations.

The comparative analysis also uncovers a transparency gap that compounds the dilemma. Washington publishes strategies without operational detail, Beijing maintains strategic ambiguity almost entirely, and Moscow treats ambiguity itself as an instrument of policy. Without some visibility into each other’s AI capabilities and intent, none of the three powers can build the confidence needed to slow competitive dynamics. The study is careful about evidence: documented incidents such as loitering munitions engaging targets in Ukraine, or the 2023 downing of a US MQ-9 Reaper over the Black Sea, illustrate operational tendencies rather than proven causal effects on strategic stability.

Crisis management may be where the risks concentrate most sharply. Historical stability rested on human judgment exercised under pressure, as during the Cuban Missile Crisis, and on communication channels that allowed signaling and de-escalation. Autonomous systems inserted into escalation chains introduce the possibility of emergent behavior, unpredictable outcomes arising when rival machine systems interact, and systemic fragility, in which tightly coupled networks cascade from a single misidentification. No such uncontrolled escalation has been reported, but the study argues that the theoretical vulnerability is qualitatively new: for the first time, actors with no capacity for judgment could occupy positions in the chain that leads to war.

On governance, the picture is bleak but not hopeless. United Nations talks on lethal autonomous weapons have stalled over great-power disagreement and the dual-use nature of AI, leaving a governance void that encourages self-help strategies. Edmond outlines plausible confidence-building measures, including advance notice of AI-enabled exercises, consultations before deploying new autonomous systems, commitments to keep humans in nuclear command and control, and transparency in testing. He also proposes expanding strategic stability to include system stability, the capacity to run tightly coupled technical networks without catastrophic failure, and attribution stability, the ability to trace actions to their source. The study’s conclusion is measured: absent effective governance, growing autonomy will likely reduce strategic stability, a likely risk rather than a certain fact, and whether the world acts on that warning now depends on governments still racing to master the technology.

Subject of Research: The impact of AI-enabled autonomous weapons on strategic stability and escalation risk among major powers

Article Title: Autonomous weapons and strategic stability

Article References: Edmond, C. (2026). Autonomous weapons and strategic stability. Discover Global Society, 4(1), Article 221. https://doi.org/10.1007/s44282-026-00548-7

Image Credits: AI Generated

DOI: 10.1007/s44282-026-00548-7

Keywords: autonomous weapons, artificial intelligence, strategic stability, security dilemma, escalation dynamics, military doctrine, deterrence, human control, United States, China, Russia, arms control