North Korea’s Military Use of Artificial Intelligence and Peace on the Korean Peninsula: 

AI, Asymmetric Strategy, and Deterrence Stability

Hee-soo Yoon

Senior Researcher

Institute for Unification and Peace

Seoul National University

From Declaration to Plan: The Legacy of the 9th Party Congress

At the 9th Congress of the Workers’ Party of Korea in February 2026, North Korea identified intelligent weapons systems—including “various AI-enabled unmanned attack complexes”—as key priorities under its five-year national defense development plan. The regime also claimed that artificial intelligence had been incorporated into the 600-millimeter multiple rocket launcher unveiled shortly before the Party Congress. These developments are difficult to dismiss as a one-time demonstration of technological capability. In 2024, Kim Jong-un ordered the integration of artificial intelligence into unmanned aerial vehicles (UAVs). During his inspection of the Unmanned Aerial Technology Complex in March 2025, he designated unmanned systems and artificial intelligence as “sectors that must be given top priority and developed” as part of the modernization of the armed forces. Since 2024, Rodong Sinmun has repeatedly cited foreign assessments characterizing artificial intelligence as the “third revolution” in military affairs, following gunpowder and nuclear weapons.

The key issue is the degree of institutional and strategic consolidation that has already taken place. As political discourse, institutional arrangements, and defense planning have advanced at different rates, the 9th Party Congress effectively formalized these developments within the framework of national planning. The fact that these ambitions have been incorporated into a formal plan suggests that they will be accompanied by resource allocation and organizational restructuring. It also means that progress over the coming years is likely to become increasingly observable.

How this development is interpreted will shape the parameters of the policy response. If North Korea’s efforts are viewed primarily as an attempt to close a technological gap, the response may focus largely on managing that gap. But North Korea’s militarization of artificial intelligence is better understood as the latest technological manifestation of a strategic logic that has evolved over decades to compensate for its structural inferiority in conventional military power. If that assessment is correct, the appropriate response must be strategic rather than merely technological.

Why AI? The Technological Extension of Asymmetric Strategy

A central principle of asymmetric warfare is that material superiority does not, by itself, determine the outcome of a conflict. A weaker state can offset an adversary’s superior capabilities by targeting its vulnerabilities, increasing the costs of military action, and employing strategies that exploit asymmetries in the way each side conducts warfare. North Korea’s development of nuclear weapons, missiles, and cyber capabilities as the Korean People’s Army’s three principal asymmetric capabilities is broadly consistent with this logic.

It is important, therefore, not to misunderstand the role of artificial intelligence in this context. For a relatively weaker state, AI does not necessarily represent an entirely new category of military capability. Rather, it functions as a “force multiplier” that can improve the effectiveness of existing asymmetric capabilities while increasing operational ambiguity. This amplification operates through at least three mechanisms.

The first is cost-effectiveness. Low-cost unmanned systems can generate favorable exchange ratios when employed against much more expensive conventional platforms. The second is greater accessibility. The proliferation of commercial drones, sensors, and open-source AI models enables states to develop capabilities in selected areas without possessing a sophisticated indigenous defense-industrial base. The third is a potential shift in the escalation threshold. AI-enabled cyber operations and unmanned aerial systems can complicate attribution and make it more difficult to determine an adversary’s intent, thereby creating greater opportunities for low-intensity activity in the gray zone.

Debates over autonomous weapons systems have traditionally focused on the capabilities and risks associated with technologically advanced states. Yet similar concerns arise when actors seek to maximize asymmetric effects by combining relatively inexpensive commercial technologies. Recent scholarship has increasingly shifted its focus from the question of “who possesses the most advanced models” to the question of “who can select and combine the most effective capabilities under resource constraints.” This dynamic is particularly relevant to North Korea, where severe resource limitations may actually encourage the selective adoption of low-cost, high-efficiency technologies.

Cyber, Unmanned Systems, and Nuclear Command and Control: Three Dimensions of Integration

In the cyber domain, North Korea maintains a substantial operational force distributed across organizations such as Unit 121 of the Reconnaissance General Bureau and the Command Automation Bureau of the General Staff Department. Its strategic objectives can broadly be understood as compensating for conventional military weaknesses, generating coercive leverage at relatively low cost, and earning foreign currency in an environment of international sanctions.

The qualitative dimension of these capabilities is changing as they become increasingly integrated with artificial intelligence. Numerous reports have documented the use of generative AI to automate phishing content, create deepfakes for identity spoofing, facilitate fraudulent remote IT employment schemes, and provide real-time AI assistance during infiltration attempts. The $1.5 billion in cryptocurrency reportedly stolen in the February 2025 Bybit hack, together with estimates that North Korean cyber operations generate approximately $2 billion annually in cryptocurrency-related theft, suggests that these activities should not be viewed simply as isolated criminal enterprises. Rather, they appear increasingly integrated into a broader system for generating revenue outside the formal international financial system—a kind of “shadow state treasury” designed in part to circumvent sanctions.

AI can further enhance the efficiency of what lies at the core of asymmetric cyber strategy: exploiting vulnerabilities. It can expand the scale of attacks, make it easier to evade detection, and increase the credibility and sophistication of social-engineering operations.

In the domain of unmanned systems, progress has continued since the 8th Party Congress in 2021, when the intelligent transformation and unmanned modernization of military equipment were identified as priorities under the Five-Year Plan for the Development of Defense Science and Technology. In March 2025, North Korean state media reported that Kim Jong-un had observed performance tests involving “AI-enabled” suicide and reconnaissance drones. Publicly released footage showed suicide drones striking simulated targets resembling South Korean L-SAM launchers, K1 tanks, and U.S. Stryker armored vehicles. This raises the possibility that some of these systems incorporate rudimentary target-recognition capabilities based on visual or shape recognition.

The asymmetric value of AI-enabled drones lies partly in the unfavorable cost-exchange ratio they can impose on the defender. Large-scale deployment of relatively inexpensive drones can force the defending side to expend substantially more expensive interceptor systems. If autonomous navigation and target recognition become feasible even in environments where satellite navigation is degraded or jammed, some of the U.S.-Korea alliance’s advantages in electronic warfare could be partially offset. North Korea’s reported acquisition of counter-drone and electronic-warfare tactics through its involvement on the Ukrainian battlefield adds another variable to this assessment.

The area with potentially the most serious strategic implications is nuclear command and control (NC2). According to an internal North Korean military directive reportedly obtained in August 2025, the General Staff instructed the Strategic Forces to develop a phased plan for an AI-based integrated management system covering the storage, deployment, and command of nuclear weapons, as well as nuclear counterstrikes.

Such a concept is rational within the logic of asymmetric deterrence. One of the principal vulnerabilities of North Korea’s nuclear posture is its potential exposure to decapitation strikes. If a retaliatory capability could be exercised without direct intervention from the leadership, the regime could strengthen its confidence in the credibility of assured retaliation. At the same time, however, such automation introduces a classic “out-of-the-loop” risk: a communications failure, sensor malfunction, or erroneous warning could be misinterpreted as evidence of a decapitation strike and potentially trigger premature nuclear use.

These three areas are closely connected to a fourth dimension: internal political control. The collection of behavioral data through the Kwangmyong Network, the deployment of facial-recognition systems at airports and railway stations, and the accumulation of biometric information across the population as North Korea moves toward electronic identification systems rely on technological foundations similar to those supporting the regime’s external intelligence and military operations.

This convergence suggests that North Korea’s AI strategy may serve a dual purpose: strengthening the regime’s internal resilience and expanding its asymmetric military capabilities externally. The same data, algorithms, and technological infrastructure can potentially support both functions.

Between Intent and Capability: Neither Overestimating nor Underestimating the Threat

Overestimating North Korea’s capabilities can itself generate policy errors. Of the three basic components of AI development—algorithms, data, and computing resources—North Korea faces particularly significant constraints in computing capacity and electricity supply. Export controls on advanced semiconductors restrict its access to high-performance computing resources, while its limited domestic ICT manufacturing capacity means that it remains dependent on external sources for servers, networking equipment, and other critical infrastructure.

As of 2020, North Korea’s total electricity generation was less than 5 percent of South Korea’s, making it difficult for the country to maintain the reliable power infrastructure required for sustained large-scale AI training. Accordingly, publicly reported North Korean AI research appears to place considerable emphasis on algorithmic optimization as a means of reducing the impact of infrastructure constraints. Examples include multi-object tracking, lightweight speech synthesis, and intonation recognition.

This approach reflects an effort to convert structural constraints into asymmetric technological adaptation. At the same time, it makes external assessment more difficult because the actual performance and operational maturity of these systems are often difficult to independently verify.

At this stage, it is therefore reasonable to characterize North Korea’s military AI capabilities as entering a phase in which strategic intent is being formalized while partial capabilities are being accumulated. Yet the incompleteness of those capabilities should not be mistaken for an absence of risk.

The very articulation of an intention to integrate AI into nuclear command and control can function as a strategic signal. It has implications for how North Korea perceives the ROK-U.S. alliance’s Kill Chain concept and can complicate allied deterrence calculations even before such systems achieve full operational deployment.

Moreover, the potential transfer of automation, deception, and unmanned-systems technologies from Russia, combined with North Korea’s ability to exploit logistics, financial, and technology channels through China and other intermediaries, could accelerate the transition from experimentation to operational capability.

The central policy challenge, therefore, is to avoid two opposing errors: underestimating North Korea’s ability to adapt emerging technologies for asymmetric purposes and overestimating the maturity of its underlying AI capabilities. The more consequential question is not whether North Korea can immediately match the most advanced AI powers. It is whether Pyongyang can selectively integrate relatively inexpensive AI-enabled technologies into its existing nuclear, cyber, and unmanned-warfare capabilities in ways that alter the strategic balance.

If the answer is yes, the implications for deterrence stability on the Korean Peninsula are substantial. The policy response should consequently focus not only on maintaining technological superiority but also on preserving crisis stability, strengthening resilience against AI-enabled attacks, establishing clear communication mechanisms, and ensuring that greater automation does not inadvertently lower the threshold for military escalation.

In this sense, North Korea’s military adoption of artificial intelligence should be understood less as an isolated technological development than as an extension of its longstanding asymmetric strategy. The central challenge for South Korea and the United States is to ensure that technological innovation does not undermine the fundamental objective of deterrence: preventing conflict while preserving the ability to manage crises before they escalate beyond control.

How Is Deterrence Being Undermined?

The challenges posed by North Korea’s integration of artificial intelligence into its military capabilities can be summarized in three distinct but mutually reinforcing ways. The most immediate challenge is the growing uncertainty surrounding attribution. If AI enables the automation of critical stages of an attack while minimizing identifiable human involvement, the “cost-benefit calculation of a rational actor”—a fundamental premise of retaliatory deterrence—can break down when attribution becomes uncertain or impossible.

The second challenge is the destabilization of the escalation ladder. When the pace of military decision-making accelerates beyond the capacity of human decision-makers to fully assess available information, the risk that an accidental or unintended incident will rapidly escalate to higher levels increases. This risk is particularly acute on the Korean Peninsula, where limited strategic depth leaves decision-makers little time to distinguish between deliberate attacks, accidents, and false alarms. Moreover, excessive reliance on advanced technologies could gradually alter perceptions of nuclear weapons, which have traditionally served primarily political and psychological deterrent functions, by making them appear more usable as military instruments. Such a development could progressively weaken the normative barriers against nuclear use.

The third challenge is a deepening asymmetry in deterrence. South Korea and the United States may be required to develop increasingly sophisticated and costly defensive capabilities, while North Korea could impose significant pressure on those capabilities through relatively inexpensive technologies. Under such conditions, the possibility of an “attacker’s advantage” cannot be dismissed. It is important to recognize, however, that these three challenges do not operate independently. Uncertainty over attribution can undermine the ability to manage escalation, while escalation risks can increase the costs of defense and thereby further deepen the asymmetry in deterrence. This structural interdependence lies at the core of the challenge posed by North Korea’s military adoption of AI.

A Timeline for Response: Short-, Medium-, and Long-Term Priorities

In the short term, the most urgent priorities are to rebalance the cost-exchange ratio and expand the range of early-warning indicators. The current model of relying on high-cost interceptor systems to counter relatively inexpensive drones is unlikely to be sustainable over the long term. South Korea and the United States should seek to lower the cost of defense through a layered combination of lower-cost interception capabilities, AI-enabled detection and identification, electronic warfare, and directed-energy systems.

At the same time, the allies should establish mechanisms for identifying emerging capabilities before they reach operational maturity. North Korea’s research and development trends in video and audio AI, advances in lightweight algorithms, and efforts to secure computing resources and critical components through alternative procurement channels could serve as early-warning indicators for adjusting sanctions enforcement, export controls, and supply-chain monitoring.

In the medium term, the analytical and policy focus should shift from individual weapons systems to the underlying functions they perform, while attribution capabilities appropriate to the AI era should be institutionalized. As long as North Korea’s AI-enabled capabilities rely on common functions across cyber operations, unmanned systems, and command and control—including target identification, automated intrusion, and decision-support functions—a response architecture divided strictly by operational domain will leave significant structural vulnerabilities.

What is required is an integrated assessment framework capable of treating these capabilities as interconnected components of a broader threat ecosystem. Moreover, effective retaliatory deterrence depends on rapid, credible, and reliable attribution. Expanding ROK-U.S. joint cyber exercises, strengthening real-time intelligence- and information-sharing mechanisms, and deepening cooperation with the private cybersecurity sector would help strengthen the technical foundations of credible deterrence.

In the long term, two broader questions will become increasingly important: South Korea’s role in shaping international norms and the degree of strategic autonomy it can realistically maintain. Although an international consensus is emerging that human judgment should not be displaced in decisions concerning the use of nuclear weapons, the more subtle and systemic risks created by integrating AI throughout command-and-control architectures have received comparatively less attention. These risks include the possibility that AI-mediated information flows could distort a national leader’s situational awareness or compress the time available for deliberation during a crisis.

South Korea therefore has an opportunity to participate in these debates not merely as a rule-taker or beneficiary of emerging norms, but as an agenda-setter. At the same time, strategic autonomy in AI should be understood realistically. For middle powers that cannot fully secure independent access to data, computing capacity, advanced semiconductors, and critical digital infrastructure, pursuing technological self-reliance can easily become another form of external dependence.

Just as North Korea is developing asymmetric capabilities while relying to varying degrees on Chinese and Russian technology, infrastructure, and procurement networks, South Korea is also embedded in—and therefore exposed to—the technological ecosystems of major powers. Strategic resilience will consequently require not isolation from those ecosystems, but greater diversification, redundancy, and the capacity to make independent decisions within them.

From Managing the Environment to Shaping It

North Korea’s militarization of AI is not simply a question of the performance of individual weapons systems. It represents a potential challenge to the underlying structure of deterrence itself. Because AI-enabled capabilities may begin to alter an adversary’s strategic calculations before they achieve full operational maturity, there is little justification for waiting until these technologies are fully deployed before developing an appropriate response.

At the same time, an excessive response based on an inflated assessment of North Korea’s capabilities could lead to inefficient resource allocation, unnecessary escalation, and an avoidable arms competition. What is required instead is a calibrated approach that continuously assesses the gap between North Korea’s strategic intentions and its actual capabilities and adjusts the intensity of the response accordingly.

More fundamentally, however, the nature of the response must evolve. A reactive, management-oriented strategy that responds to technological developments only after they emerge will struggle to address the next generation of capabilities developing in the gray zone between conventional conflict and peacetime competition.

South Korea should therefore move beyond simply managing the technological environment created by others. It should acknowledge the realities of external technological dependence while developing its own clearly defined response frameworks, operational guidelines, and long-term roadmaps. At the same time, Seoul should seek to contribute actively to the development of international norms governing deterrence, crisis management, human control, and the responsible use of AI in military affairs.

Only by combining technological resilience with strategic initiative can South Korea move from managing the emerging security environment to helping shape it.