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

The theoretical foundation of the analysis rests on the classical triad of deterrence: capability, credibility and communication. Capability means survivable second-strike forces, typically a triad of land-based intercontinental ballistic missiles, submarine-launched ballistic missiles and strategic bombers. Credibility exists when an adversary believes retaliation will actually follow if a red line is crossed, a belief shaped by doctrine, political will and alliance obligations. Communication involves signaling what behavior is unacceptable while preserving useful ambiguity. During the Cold War, mutual assured destruction anchored these three elements in a bipolar standoff with clearly constructed escalation ladders. Today, the study argues, asymmetrical arsenals, emerging technologies and cross-domain interactions have stretched deterrence far beyond simple mutual assured destruction toward concepts of flexible response and tailored deterrence.

Hybrid warfare, as defined in the review following Frank Hoffman’s framework, is the coordinated use of conventional force, cyber operations, disinformation, economic coercion, proxy forces and legal manipulation, or lawfare, to achieve strategic objectives without formally declaring war. Its subset, gray-zone activity, deliberately blurs attribution, escalation and the distinction between state and non-state actors. The author distinguishes the two with concrete examples: Russia’s 2014 annexation of Crimea involved hybrid tools such as deniable forces and cyber-attacks but crossed into conventional military operations once troops entered the peninsula, whereas the 2017 NotPetya cyber-attack achieved destructive effect without any physical incursion, remaining firmly in the gray zone. What makes contemporary hybrid warfare distinctive is not novelty, since propaganda and proxy wars are ancient instruments of statecraft, but the speed and coordination that digital technologies enable.

The central theoretical claim is that nuclear deterrence casts a shadow under which hybrid competition flourishes through three mechanisms. First, the perceived risks of escalation prevent direct military confrontation between nuclear powers. Second, a credibility gap opens between nuclear threats and hybrid actions: threatening nuclear retaliation against a cyber intrusion or an election interference campaign appears disproportionate and therefore incredible. Third, the cumulative effect of many individually minor sub-threshold actions can shift regional balances of power without ever triggering escalation. This is a modern extension of the stability-instability paradox first articulated by Glenn Snyder, which held that strategic nuclear stability permits lower levels of conflict. The review argues the paradox has become more pronounced in the hybrid era, and that adversaries increasingly operate in realms where nuclear threats are neither credible nor proportionate.

The empirical cases are drawn from Ukraine, NATO’s eastern flank and the Indo-Pacific. The war in Ukraine is presented as the most demanding test: nuclear deterrence did not prevent Russia’s conventional invasion of February 2022, but it has constrained direct combat between Russia and NATO. Russian nuclear signaling, including public references to strategic forces, has sought to limit Western intervention, while NATO members have supplied Ukraine with military, financial and intelligence support without direct intervention, demonstrating calibrated escalation management. The absence of direct NATO-Russia combat suggests deterrence remains structurally intact, while the ongoing large-scale conventional war shows it cannot stop sub-nuclear conflict when core interests are perceived to be at stake. The study also notes that Western sanctions, cyber activity and intelligence sharing constitute hybrid instruments in their own right, underscoring the author’s deliberately symmetrical treatment of all major powers.

The United States case illustrates how a nuclear umbrella enables cyber coercion. The Stuxnet malware, developed jointly with Israel and deployed against Iranian nuclear centrifuges between 2009 and 2010, physically destroyed roughly a thousand centrifuges and set back Iran’s program by an estimated eighteen to twenty-four months without prompting an overt military response. The review argues Iran’s restraint reflected an assessment that escalating beyond a cyber operation under the American nuclear umbrella was unacceptable. The 2022 US Nuclear Posture Review explicitly identifies strategic competition below the nuclear threshold as the principal threat, and the doctrine of tailored deterrence seeks response options across the full spectrum of conflict, though critics such as James Acton warn that blurring conventional and nuclear distinctions raises the risk of miscalculation.

Russia, holding roughly 5,580 warheads with 1,674 deployed on strategic delivery systems, explicitly frames nuclear weapons in its doctrine as means of escalation management, and its 2020 nuclear deterrence principles widened the conditions under which nuclear use might be considered. This ambiguity raises escalation thresholds in the adversary’s mind while Moscow conducts conventional and hybrid operations beneath them. The NotPetya attack, attributed to Russian military intelligence by US, British and Dutch agencies and causing more than ten billion dollars in global damage, drew sanctions and diplomatic expulsions but no military response, which the review interprets as evidence that nuclear deterrence shields even the most destructive hybrid operations from conventional retaliation. China presents a different profile: a rapidly expanding arsenal projected to grow from roughly 350 warheads in 2020 to nearly 1,500 by 2035, a declared no-first-use policy, and gray-zone coercion through maritime militia operations such as the 2012 Scarborough Shoal confrontation, where state-coordinated civilian vessels denied Philippine access without triggering a military response.

Several structural findings sharpen the analysis. Attribution ambiguity, the inability to identify a hybrid attacker in near real time, undermines deterrent signaling because threats cannot credibly be directed at an unconfirmed actor. Cumulative coercion aggregates sub-threshold incidents, each too minor to justify retaliation, into a strategic shift. Perhaps most dangerously, cyber operations against early-warning systems, command-and-control networks or dual-use satellites create entanglement risks: if a state believes its nuclear deterrent is compromised, it may adopt more aggressive launch postures, raising the probability of escalation through miscalculation rather than intent. Meanwhile, the collapse of the arms control architecture, including the 2002 withdrawal from the ABM Treaty, the 2019 demise of the INF Treaty and the expiration of New START in 2026, has removed verification mechanisms, transparency and crisis communication channels that once reduced the likelihood of miscalculation.

The review also challenges the assumption that democracies are structurally constrained in hybrid competition. The 2003 invasion of Iraq, launched despite mass public opposition and without UN Security Council authorization, and covert cyber and drone operations conducted with limited accountability demonstrate that democratic institutions shape but do not prevent coercive action. Conversely, authoritarian regimes face their own constraints: Russia must manage coalitions with China and India, while China balances regional assertiveness against economic stability. The author concludes that the democracy-autocracy distinction is an analytical failure for assessing nuclear signaling and hybrid behavior, and that credibility is better evaluated through strategic culture, alliance commitments and domestic political structure. Credibility itself has become a multi-domain, dynamic property, resting not only on warhead counts but on political coherence, narrative control and alliance cohesion, all of which hybrid disinformation campaigns deliberately target.

The policy prescriptions follow directly from the diagnosis. Rebuilding strategic stability requires renewed arms control, even limited agreements on notification and data exchange, plus confidence-building measures against cyber interference in nuclear command systems. Deterrence by denial deserves priority at sub-nuclear levels: cyber resilience, electoral protection, critical infrastructure hardening, information literacy and economic diversification all raise the cost of gray-zone operations without escalating risk. Crisis communication channels, transparency in military exercises, and international norms prohibiting cyber interference with nuclear command structures would reduce the danger of inadvertent escalation. The study proposes an integrated cross-domain model combining nuclear credibility, political legitimacy, alliance cohesion and societal resilience. Its conclusion is sobering but precise: nuclear deterrence is neither obsolete nor absolute. It remains structurally powerful as a ceiling on total war, yet operationally constrained beneath that ceiling, and the struggle for influence among the great powers will increasingly be decided in the ambiguous, cumulative and deniable forms of competition that flourish in its shadow.

Subject of Research: The effectiveness of nuclear deterrence against hybrid and gray-zone warfare among major powers

Article Title: Nuclear deterrence in the age of hybrid warfare

Article References: Edmond, C. (2026). Nuclear deterrence in the age of hybrid warfare. Discover Global Society, 4(1), Article 228. https://doi.org/10.1007/s44282-026-00419-1

Image Credits: AI Generated

DOI: 10.1007/s44282-026-00419-1

Keywords: nuclear deterrence, hybrid warfare, gray zone, strategic stability, stability-instability paradox, cross-domain deterrence, arms control, escalation, cyber operations, Russia, China, NATO