• Online Exclusive Article Archive

By Algorithm or Order

Integrating Lethal Autonomous Weapon Systems into Targeting

 

Maj. Michael J. Brodka, US Army

 

Download the PDF

 

On a distant battlefield in the near future, machine signals govern action where human voices once commanded. Swarms of autonomous, unmanned systems patrol the battlespace, linked through distributed AI and mesh networks. These systems collect and fuse data from across domains to rapidly detect enemy formations. Within moments, a real-time common intelligence picture emerges, mapping US Army forces. What once required hours of analysis now occurs in seconds.

Adversarial AI produces false signatures to manipulate US radars and inject phantom objects into intelligence, surveillance, and reconnaissance (ISR) feeds. Concurrently, autonomous ISR identifies US radars, logistics nodes, and communications equipment and then transmits their locations via the mesh network. Across the brigade combat team, sensors, shooters, and command posts are struck simultaneously by integrated fires and drone swarms. US combat power disintegrates within minutes. The battle is decided by algorithm, not by order.

This operational environment is not drawn from science fiction but instead reflects the natural evolution of warfare. War endures—passionate, volatile, and political by its nature, as Carl von Clausewitz observed—but its character is never fixed.¹ War takes the shape of the societies that fight it and the technologies they employ. In one era, it maneuvers with horse and saber; in another, it fights through code and circuit. As AI and machine learning proliferate, the character of war accelerates toward an algorithmic future. China’s concept of intelligentized warfare and Russia’s reconnaissance-strike complex (RSC) point toward a battlespace increasingly defined by speed and autonomy.² In this environment, adversaries are preparing to fight by algorithm, treating autonomy itself as a competitive advantage over human cognition.

One indicator of this shift is the emergence of lethal autonomous weapon systems (LAWS), which can converge sensing, targeting, and engagement to near simultaneity. Both Beijing and Moscow view AI-enabled autonomy as central to future warfighting.³ In China, the People’s Liberation Army (PLA) seeks cognitive dominance by shifting the human role from continuous input to supervisory control, using machine speed to offset numerical or material disadvantages.⁴ Russia, by contrast, pursues automated command-and-control (C2) architectures designed to fuse real-time sensing with precision strike.⁵ Although neither state can yet field fully autonomous weapons, both are designing doctrine and systems to do so. These trajectories raise a fundamental question for the Army: Can its existing doctrinal processes operate effectively in an operational environment increasingly shaped by autonomy?

The Department of Defense (DOD) has directed the Army to prepare for the employment of LAWS.⁶ This article argues that the Army’s doctrinal processes, particularly its targeting methodology, are misaligned with that requirement and are increasingly unable to keep pace as algorithmic warfare emerges. Advances in automated sensing and data fusion have dramatically compressed the time between detection and engagement. Army doctrine, however, remains anchored in human-paced decision-making. The result is a velocity gap between the speed at which information is collected and the tempo at which it can be acted upon. Adversaries are actively designing autonomous systems and supporting doctrine to exploit this gap. Closing it requires redesigning Army doctrine, organizations, and training to integrate LAWS into targeting under lawful human-on-the-loop control.

Clarifying Human Control in Lethal Autonomy

Under DOD Directive 3000.09, Autonomy in Weapon Systems, LAWS are weapons that, once activated, can select and engage targets without further human operator intervention within commander-approved constraints.⁷ The directive establishes DOD policy for the development and employment of LAWS.⁸ It requires that appropriate human judgment be incorporated into the design, authorization, and supervision of weapon-system behavior.⁹ The directive allows autonomous systems to execute engagements within predefined operational constraints consistent with commander intent.¹⁰ It also requires commanders to supervise the employment of LAWS and retain the ability to intervene or terminate operations.¹¹

Army targeting doctrine already embeds human judgment through commander-approved guidance, including target selection standards and the attack guidance matrix, which translate commander intent and legal determinations into conditions governing the use of lethal force. The question, therefore, is not whether human judgment exists in targeting, but whether that judgment can be expressed in a form LAWS can execute, and commanders can supervise.

The DOD Law of War Manual clarifies that law of armed conflict (LOAC) compliance is assessed at the time a commander authorizes an attack, based on the information reasonably available at that moment.¹² Army targeting doctrine operationalizes this judgment through preapproved guidance intended to govern execution without renewed legal deliberation at each engagement.¹³

When employing LAWS, that same commander-approved judgment must be translated into engagement logic, the parameters that define how autonomous systems operate once authorized. Without this translation, LAWS cannot operate at machine speed, thereby weakening commander control and undermining lawful execution.

The result is a misalignment between what policy permits and what doctrine enables. Policy allows commanders to authorize autonomous execution, provided they can supervise and intervene as required. Army targeting doctrine, however, provides no mechanism to encode the commander’s judgment for autonomous execution. This leaves the Army bound to human-paced processes, even as adversaries pursue machine-speed advantage through AI-enabled autonomy.

Adversarial Trajectories: Toward Algorithmic Advantage

This asymmetry in control models carries operational consequences that adversaries understand well. Moscow and Beijing are not waiting for the US to resolve how LAWS should be integrated. They are already moving to exploit automation itself as a source of advantage. As Elsa Kania observes in PRISM, “Algorithmic advantage may become a dominant determinant of operational advantage,” particularly for militaries seeking to offset conventional disadvantages.¹⁴ For China, the contest for cognitive superiority through algorithmic advantage is not a supporting effort in future war; it is the terrain on which operational advantage is expected to be decided. Russia, meanwhile, pursues a similar objective through the continued evolution of its RSC, aiming to compress sensor-to-shooter timelines toward near-real-time execution.

Russia’s algorithmic lineage. Russia’s pursuit of algorithmic warfare reflects a long theoretical lineage rather than a modern technological evolution. Soviet theorists developed this tradition through experience with industrialized war, recognizing that conflict unfolded through overlapping operations with effects that accumulated over time. Aleksandr Svechin emphasized campaigns built through linked operations rather than decisive engagements.¹⁵ Georgii Isserson pushed the logic further, arguing that war was defined by depth, simultaneity, and the disruption of an opponent’s system.¹⁶ War, in this view, became something to be shaped over time rather than decided in a single moment.

Long-range fires, networked sensors, and automated processing started to compress time and distance in ways that could collapse the gap between conventional and nuclear effects.¹⁷ Marshal Nikolai Ogarkov recognized this and focused on fusing these capabilities to strike deep, quickly, and with mass.¹⁸ From that logic emerged what would later be described as the RSC: a fires architecture built to decide, detect, and deliver faster than an adversary could respond. Contemporary analysts noted that the Soviets “may even be ahead of the West” in recognizing automation’s battlefield implications.¹⁹

These ideas did not disappear after the collapse of the Soviet Union. They reemerged as Russia rebuilt its forces around precision strike and automated C2. By the early 2010s, senior planners were again focused on how information technologies were compressing time and distance on the battlefield. Russian Chief of the General Staff Valery Gerasimov later captured this shift in 2016, describing a form of warfare defined by “long-distance, contactless actions” and the growing role of robotic systems.²⁰ Gerasimov’s direction was clear: Future formations would rely increasingly on automation rather than on human-paced control.²¹

The war in Ukraine has demonstrated this lineage in practice. An analysis by the Royal United Services Institute describes Russian fire-control systems, such as Strelets, as integrating multiple sensor feeds into digital architectures that support tiered RSCs across echelons. Rather than linear kill chains, Russian forces increasingly employ interconnected “reconnaissance-fire circuits.”²² Lancet loitering munitions often serve as the kinetic endpoint of this process, with engagements occurring within minutes of target detection. These exchanges between Ukraine and Russia have become a “competitive duel,” where the side that identifies targets first often kills first.²³

Chinese cognitive control. China’s trajectory differs in its methods but not in its strategic intent. Where Russia’s approach reflects the gradual mechanization of a scientific theory of war, China’s concept of intelligentized warfare represents a deliberate effort to redesign conflict around cognitive dominance. This, the PLA believes, is achieved through human–machine teaming. In PLA doctrinal writing, intelligentized warfare is not defined by autonomous platforms alone but by the system-level orchestration of sensing, cognition, and action to dominate an adversary’s decision-making process.²⁴

This approach is rooted in ideas that have long shaped Chinese military thinking, particularly the emphasis on foreknowledge, deception, and advantage gained through control of information. Those themes recur in the writings of Sun Tzu and Mao Zedong to shape an adversary’s decisions before force is applied. Analyses by the US Army Foreign Military Studies Office affirm that contemporary PLA concepts retain this logic while adapting it to modern technology.²⁵ These ideas continue to inform Chinese operational thinking. Official doctrinal texts such as Science of Military Strategy emphasize information dominance and psychological effects in informatized warfare.²⁶ Later PLA writings on intelligentized warfare increasingly extend competition into the cognitive domain through public opinion struggle, psychological offense and defense, and cognitive opposition.²⁷

Building on this foundation, the PLA frames intelligentized warfare as a system-of-systems approach designed to compress observation, decision, and action through algorithmically enabled processes. Chinese doctrinal writing contrasts this approach with what it characterizes as a US reliance on mass and scale, arguing instead that tempo dominance can defeat size.²⁸ Within this logic, autonomous reconnaissance and human-machine teaming are intended to operate faster than human cognition alone, generating a decisive advantage in targeting.²⁹

To achieve this advantage, the PLA does not treat AI as an adjunct to planning and C2. Instead, AI is considered C2’s cognitive backbone. Commanders provide intent and constraints, while autonomous systems execute detection, prioritization, and sequencing.³⁰ By reducing friction and compressing decision cycles, speed itself becomes a source of coercive pressure. Human judgment does not disappear; it shifts earlier in the process, shaping execution through supervision rather than by pacing individual actions. The intent is not automation alone but a reorganization of warfare around algorithmic coordination.³¹

The Velocity Gap

The velocity gap is the operational manifestation of this asymmetry between machine-tempo execution and human-paced control. Russia’s RSC and China’s intelligentized warfare models are designed to integrate decision-making, sensing, and strike into near-simultaneous execution via autonomous systems. In contrast, the Army’s targeting enterprise remains structured around sequential human validation and approval across the targeting process. As a result, the limiting factor is not sensing or computation, but the processes that govern action.

Recent experimentation demonstrates that AI-enabled tools and autonomy can significantly compress portions of the kill chain, but only to the extent permitted by human-in-the-loop doctrine. The 18th Airborne Corps reduced sensor-to-shooter timelines from 724 minutes to approximately twenty minutes using Project Maven, with similar results observed by the 4th Infantry Division during Next Generation C2 (NGC2) testing at Ivy Sting.³² Both results confirm that technology can accelerate sensing and shooter cueing; however, in each instance, action proceeded at the pace of human processes during target prosecution.³³

Brigade-level field experimentation at the Joint Readiness Training Center in the summer of 2024 reinforces this conclusion. Operational trials within the 2nd Brigade, 101st Airborne Division, demonstrated that machine-learning tools significantly improved tactical reconnaissance by accelerating object detection from drone-collected full-motion video.³⁴ These systems were employed strictly as collection tools to enhance situational awareness rather than to select or engage targets. Although the experiment did not address targeting, it reflects the same structural pattern observed elsewhere: AI accelerates sensing and information processing, while execution remains governed by human-paced doctrinal processes.

In this context, velocity is not simply a measure of speed; it determines when human judgment can be exercised. As sensing, targeting, and engagement converge toward simultaneity, judgments that doctrine assumes will occur proximately to execution must instead happen earlier. If they are not, human decision-making cannot keep pace with autonomy. The result is not faster or better judgment but a forced choice between operational paralysis, as approvals lag behind events, and premature risk acceptance, as commanders authorize actions without full situational awareness.

This dynamic explains why the resulting gap cannot be closed through materiel or organizational fixes alone. Army Doctrine Publication (ADP) 1-01, Doctrine Primer, notes that the impact of changes across doctrine, organization, training, materiel, leadership and education, personnel, and facilities “cannot be fully realized without a significant change in doctrine.”³⁵ When doctrine falls behind, material and organizational solutions can only mitigate the problem. For that reason, the Army’s current targeting approach reflects a doctrinal assumption that will become increasingly consequential as autonomous capabilities mature. The following section analyzes three doctrinal shortfalls that prevent the Army from integrating LAWS: tempo, execution authority, and assessment and control.

Doctrinal Shortfalls: Tempo, Authority, and Control

ADP 3-13, Information, makes the tempo competition explicit, noting that “the force that anticipates better, thinks more clearly, decides and acts more quickly, and adapts more rapidly” is best positioned to seize and retain the initiative.³⁶ Yet Army doctrine has not translated this recognition into a targeting architecture capable of operating at machine tempo. The publication that explains why information advantage matters (ADP 3-13) and the one that governs how the Army targets (Field Manual [FM] 3-60, Army Targeting) are not aligned. Autonomy is acknowledged in theory but not operationalized in practice.

Tempo. FM 3-0, Operations, introduces the first friction point in Army doctrine: a tempo shortfall. It defines the operations process (planning, preparing, executing, and assessing) as the engine for Army activities, but that process is human paced. The military decision-making process (MDMP) is deliberate and sequential, with targeting integrated throughout. However, current doctrine offers no procedural framework for employing AI-enabled decision-support tools, algorithmic engagement logic, or autonomy within the integrating processes. FM 3-60 and FM 5-0, Planning and Orders Production, likewise constrain targeting and decision-making to human-paced cycles. This creates an inherent contradiction. FM 3-0 requires a tempo advantage in multidomain operations but prescribes only human-paced processes to achieve it.

Decide, detect, deliver, and assess (D3A) is optimized for kill chains in which sequential sensor-to-shooter action is executed with continuous human input. Although D3A is described in FM 3-60 as flexible and iterative, it implicitly assumes that human judgment remains the pacing function across all four phases.³⁷ D3A requires human validation, authorization of engagements, and assessment of results, rather than providing a framework to express commander attack guidance as preapproved LAWS engagement logic.

This structure breaks down under machine-tempo conditions. As sensing and strike converge into one process, D3A concentrates human judgment precisely at the point at which time is least available. The result is not only slower execution but also a doctrinal bottleneck that prevents autonomous systems from operating at their designed pace. Rather than enabling lawful human-on-the-loop autonomy, D3A reintroduces human-in-the-loop latency into every engagement, re-creating the velocity gap it is meant to close.

Execution authority. Army doctrine further assumes that command authority is exercised close to execution. Planning doctrine reinforces this assumption through delegated authorities designed to preserve command authority when conditions change or when access to the commander is limited. FM 5-0 treats execution authority as an extension of the commander’s intent, expressed through delegated decision authorities established during mission analysis and codified in decision-support and delegated-authority matrices. These tools define who may act and under what conditions when anticipated decisions arise. They preserve command authority during execution, but they presume continuous human judgment as conditions evolve.³⁸

Targeting doctrine reflects the same logic. FM 3-60 carries the commander’s judgment forward through preapproved targeting guidance. This framework, however, assumes that execution is performed by human operators exercising judgment at the point of action rather than by autonomous systems that operate independently once authorized. In a machine-tempo environment, that assumption no longer holds. Command authority must persist through execution in a form that allows LAWS to apply it independently. DOD Directive 3000.09 explicitly permits this.³⁹

Army doctrine provides no mechanism for encoding that authority in executable form. As a result, autonomy is not excluded from doctrine, but it remains doctrinally ungoverned during execution. This gap leaves commanders with an untenable choice. They can retain control only by withholding delegated execution authority, preserving human decision ownership at the cost of autonomous employment. Or they can accept autonomous execution without a doctrinal mechanism for expressing authority, undermining control, accountability, and legal responsibility.

Assessment and control. The third doctrinal shortfall is assessment and control. Current doctrinal models assume that the conditions under which lethal force is authorized remain stable through execution. Once execution is delegated to autonomous systems, that relationship breaks down. Assessment can no longer focus solely on post-engagement effects, and control can no longer be limited to human direction at the moment of execution. Both must instead ensure that engagement logic continues to reflect lawful human authorization as information, environments, and system performance change.

The DOD Law of War Manual requires the adoption of feasible precautions in the planning and conduct of attacks to reduce the risk of harm to protected persons and objects, including verification that targets are military objectives.⁴⁰ These obligations presume the ability to determine whether the conditions under which lethal force was authorized remain valid when force is applied. Yet Army doctrinal assessment remains effects centric. FM 3-60 focuses on battle damage, functional damage, and munitions effectiveness.⁴¹ FM 5-0’s measures of performance and effectiveness similarly presume human-directed execution and are not designed for LAWS.⁴²

These tools can describe what was struck and the effects produced. They do not evaluate whether autonomous systems remain within LOAC boundaries, whether identification confidence remained valid throughout execution, or whether system degradation invalidated prior authorization. Nor do they provide a mechanism for sustained human-on-the-loop supervision capable of detecting those conditions and halting execution when legal thresholds are exceeded.⁴³ The result is a framework that reports outcomes without verifying that autonomous lethal force remains governed by lawful human judgment under changing conditions.

DOD Directive 3000.09 addresses this gap by requiring that LAWS behave predictably and reliably, retain traceable and auditable decision logic, and be capable of termination when required. It further requires that operators maintain meaningful insight into system behavior through human–machine interfaces.⁴⁴ These requirements assume an operational architecture capable of detecting degraded performance, invalid assumptions, and conditions requiring intervention during execution.

FM 3-84, Legal Support to Operations, presumes that staff judge advocates advise commanders on LOAC compliance before targets are prosecuted, while commanders retain responsibility for each engagement.⁴⁵ In an autonomous execution environment, commanders remain legally responsible for outcomes while lacking a practical means to influence decisions once execution begins.

Current targeting doctrine fulfills this function through episodic reassessment before and after engagements, but it does not extend reassessment authority into autonomous execution once lethal action has begun.⁴⁶ Commanders must therefore encode LOAC and rules of engagement (ROE) constraints into engagement logic before execution. LOAC continues to govern the battlefield, but machine-tempo autonomy requires that legal compliance be embedded in rules and algorithms rather than enforced through human-triggered engagements.

The core problem, then, is not whether autonomous systems can be reviewed or tested in advance but whether lawful control can be maintained once they are deployed in combat. Existing assessment frameworks assume that legality can be assured through predeployment review and post-engagement effects analysis. As Annemarie Vazquez warns in Military Law Review, “Rigorous testing will ferret out many of the problems, [but] they should not be the only safeguards against the unique LOAC issues posed by autonomy in weapon systems.”⁴⁷ Autonomous execution introduces uncertainty, degradation, and emergent behavior that cannot be fully captured by weapons reviews or effects-based assessment alone. Without doctrinal mechanisms to supervise autonomous actions as conditions evolve, commanders remain legally responsible for outcomes they lack the means to govern.

Recommendations: Integrating LAWS into the Targeting Enterprise

As Vazquez argues, LOAC review processes are “ill-suited to the unique nature of autonomous weapons” because the legally decisive questions shift from the moment of trigger-pull to the moment of design.⁴⁸ Closing the velocity gap requires more than faster sensors or improved networks; it demands doctrinal, organizational, and training reforms that allow the Army to employ LAWS. The following recommendations align directly with the three doctrinal shortfalls identified in this article and provide a coherent pathway to integrate human-on-the-loop LAWS into the targeting enterprise.

Doctrinal reform: Develop an autonomous targeting methodology. This recommendation establishes an updated doctrinal targeting methodology for integrating LAWS. Subsequent recommendations will address the assessment, control, and legal review mechanisms necessary to operationalize it. The Army should update FM 3-60 to incorporate a revised version of the D3A targeting methodology designed for autonomy. This new doctrinal variant is called human decide-machine detect-machine deliver-machine assess (HDM3; see figure 1). HDM3 should align the targeting methodology to preserve human judgment during the decide phase while allowing LAWS to perform detect and deliver functions and to generate initial assessment data. Under this approach, commanders retain exclusive responsibility for validating targets and approving the target selection standards and the attack guidance matrix during the targeting coordination board (TCB). That information is then encoded in engagement logic that delegates execution authority to LAWS within commander-approved constraints for targets on the high-payoff and high-value target lists.⁴⁹

This methodology enables what this article refers to as kill webs: nonlinear targeting architectures in which distributed sensors and shooters are dynamically linked to execute engagements within commander-approved constraints, rather than through fixed, sequential kill chains. In a kill web, targeting functions are not bound to a single sensor–shooter pairing or to a linear process; any authorized sensor may cue any authorized shooter within engagement logic approved by the commander. This architectural flexibility allows targeting to adapt to disruption, degradation, and opportunity. In a linear kill chain, the loss of a single node terminates the engagement. In a kill web, the loss of a node triggers dynamic reconfiguration, with the remaining sensors and shooters reallocating to complete the mission in accordance with approved engagement logic.

Engagement logic is implemented using commander-approved parameters that govern autonomous action. These parameters specify permissible target classes, required confidence thresholds for target classification, geospatial and temporal boundaries, collateral damage limits (including no-strike entities), and clear termination criteria (see figure 2).⁵⁰ Together, they define the conditions under which LAWS may act once authorized. Doctrinally, these parameters delegate execution authority, allowing commander intent and legal judgment to persist through autonomous execution.

Engagement logic operationalizes ROE by translating authorities, prohibitions, and positive identification requirements into machine-executable constraints. Confidence thresholds, in particular, function as an explicit expression of a commander’s risk tolerance, determining the level of certainty required before lethal force may be applied. In this way, legal judgment and commander’s intent are not exercised just before the engagement; they are deliberately encoded in advance and carried forward into autonomous execution.

While engagement logic is activated during TCB, most of its content must be developed earlier in the MDMP. Enemy equipment recognition standards, no-strike entities, ROE constraints, and baseline LOAC parameters should be established during mission analysis and refined during course-of-action development, allowing the TCB to focus on contextual adjustments and authorization rather than logic construction.

Termination criteria specify the conditions under which autonomous execution must pause or abort, such as loss of communication or degradation of confidence. Model drift detection and performance monitoring allow human supervisors to judge whether LAWS remain reliable under changing and uncertain conditions.⁵¹ These mechanisms embed human-on-the-loop control through judgment within the structure of bound autonomous action.⁵² Lethal force is therefore applied only within parameters deliberately approved in advance by accountable human commanders.

By implementing HDM3 methodology, the Army can accelerate its kill webs by employing LAWS as sensor, shooter, and initial assessor. HDM3 enables rapid massing of effects within the kill web and autonomously redistributes assets to the next priority. Redundancy is built into the engagement logic, reducing supervisory burden without eliminating oversight. The phase requiring sustained human input is assessment. While LAWS can generate initial assessment data, meaningful judgment is still necessary to interpret effects and evaluate autonomous behavior. LAWS provides data while humans interpret it using input from other collectors.

HDM3 reframes human control for a battlespace defined by machine tempo without altering command authority by shifting judgment to the deliberate design of engagement logic derived from TCB decisions.⁵³ This approach accepts that future warfare will outpace continuous human input but does not negate the requirement for legal and ethical judgment. By adapting HDM3, FM 3-60 can equip commanders with tools to integrate LAWS into targeting in an environment in which “the speed of modern warfare will soon exceed the speed at which humans can orient, understand, and act.”⁵⁴

Doctrinal reform: Assessing and supervising autonomous systems. FM 3-60’s assessment framework remains oriented around battle damage assessment and effects-based measures that presume human-centric operations.⁵⁵ LAWS require doctrinal mechanisms to assess not only effects but also the reliability, confidence, and behavior of the decision processes enacted by autonomous systems. FM 3-60 should be updated to recognize machine-conducted initial assessment data as a doctrinal function, with humans responsible for validation, anomaly detection, and adjustment of engagement logic.⁵⁶

In this framework, supervision is exercised through continuous human evaluation of autonomous system behavior and performance, with commanders and staff empowered to modify or suspend engagement logic when assessment indicates degraded reliability or altered conditions. This supervisory function ensures that delegated execution authority remains valid throughout execution and is withdrawn when the conditions under which it was granted no longer hold. By shifting assessment from effects alone to include algorithmic performance, doctrine preserves meaningful human judgment while enabling lawful human-on-the-loop employment of LAWS.⁵⁷

Doctrinal reform: LOAC review for engagement-logic design. FM 3-84 embeds staff judge advocates in the targeting process to review and advise on lethal decisions before execution, assuming a human-paced model in which legal review occurs proximate to the authorization of engagements.⁵⁸ This assumption holds when commanders have sufficient time to assess targets and intervene before force is applied. However, LAWS collapses that decision space. To preserve lawful human judgment under such conditions, LOAC compliance must be exercised primarily through engagement-logic design rather than engagement-proximate review.

FM 3-84 should be updated to require staff judge advocates to advise commanders on the review of target classes, confidence thresholds, geospatial limits, collateral damage constraints, and termination criteria during the TCB.⁵⁹ Commanders remain legally responsible for the use of force by delegating execution authority through approved engagement logic, whereas LAWS execute engagements only within those parameters. This shift enables LAWS employment while maintaining appropriate human judgment and clear accountability for lethal decisions.

Organizational reform: Establishing autonomous targeting cells. To supervise HDM3 at the tactical level, the Army should establish autonomous targeting cells (ATC) within existing division and corps targeting cells. ATCs would reorganize current targeting functions to expand responsibilities for supervisory control, assessment, and adjustment of autonomous engagement logic. ATCs would

• supervise control of autonomous systems;
• monitor confidence scores, model drift, and electromagnetic warfare (EW) degradation;
• adjust or revoke delegated execution authority; and
• produce algorithmic assessment reports.

ATCs serve as the organizational counterpart to doctrinal reform, providing commanders with a dedicated node for human-on-the-loop oversight and enabling LAWS integration into targeting without sacrificing control or accountability. ATCs should be staffed by crossfunctional teams drawn from existing Army specialties, including operations research and systems analysis, field artillery, military intelligence, cyber and electromagnetic warfare, and staff judge advocates.

Operations research and systems analysis provides the analytical expertise needed to interpret algorithmic performance, confidence metrics, and degradation indicators, translating LAWS outputs into assessments that commanders can act on. Field artillery and intelligence personnel ensure alignment with targeting priorities and threat assessment. Cyber and EW personnel assess the effects of electromagnetic interference, deception, and data corruption on autonomous performance. Staff judge advocates support the review and modification of engagement logic to ensure continued compliance with LOAC and ROE.

Training reform: Institutionalizing autonomy literacy. Human judgment within LAWS targeting must be institutionalized, not improvised. Integrating HDM3 into the targeting enterprise places new cognitive, technical, and legal demands on soldiers and leaders. Supervising LAWS requires more than familiarity with the platform. It requires the ability to design engagement logic, interpret algorithmic performance, detect degradation, and intervene when autonomous behavior approaches legal or operational limits. Those processes should be institutionalized through a deliberate combination of initial entry training (IET), professional military education (PME), specialized training, and collective training.

Progressive education from IET to PME. Autonomy literacy must be embedded across the Army’s IET and PME system to ensure commanders, staffs, and soldiers understand how to employ and supervise LAWS. This education should progress with rank and responsibility, moving deliberately from conceptual familiarity to doctrinal application and operational design.

At the entry level, training should introduce the fundamentals of AI and basic autonomy concepts. Here, the Army establishes a common vocabulary and understanding of how autonomous systems sense, classify, and act. As officers and noncommissioned officers advance, PME should increasingly emphasize the application of autonomy within the targeting enterprise, including autonomous sensors, signature management, confidence thresholds, and risk-informed employment decisions. At the intermediate and senior levels, education should focus on LAWS planning and supervision for division and corps staff, including HDM3 integration, engagement-logic design, incorporating LAWS into battle rhythm events (e.g., collection management and target working groups), and legal and ethical oversight.

Autonomous Reconnaissance and Strike Course. The Army should establish an Autonomous Reconnaissance and Strike (ARAS) Course to institutionalize the supervision of LAWS within the targeting enterprise. ARAS should serve as the primary qualification course for personnel assigned to ATCs. The course should provide a functional, cross-disciplinary program tailored for fires and intelligence personnel, focused on the practical supervision of autonomous systems. ARAS should train officers, warrant officers, and noncommissioned officers and award an additional skill identifier. Graduates will be qualified to incorporate autonomy in MDMP, design and supervise delegated execution authority under HDM3, and conduct engagement-logic design and algorithmic assessments.

Collective training for LAWS employment. Consistent with the FM 7-0, Training, requirement for realistic, contested training environments, individual training must be supplemented by collective training to prepare units to employ LAWS. Collective training environments at home station and at Army combat training centers should incorporate, at a minimum,

• simulated autonomous sensors with realistic detection, misclassification, and latency profiles;
• adversarial EW and deception that affect LAWS’ confidence scores;
• machine-tempo strike sequences executed within HDM3 methodology and kill webs;
• ATC integration during command post exercises; and
• simulation of engagement logic failures and abort conditions.

These enhancements allow units to practice supervisory control, legal oversight, and rapid adjustment of engagement parameters under realistic conditions. They close the gap between conceptual understanding and operational use. Such environments are essential not only for tactical proficiency but also for preserving human judgment, as commanders and staff learn to recognize when autonomous behavior approaches legal, ethical, or reliability limits in combat.

Conclusion

The Army stands at an inflection point. War’s nature endures, but its character is accelerating toward a tempo that human cognition alone cannot match. Adversaries are not waiting for the United States to reconcile LAWS integration; they are building architectures designed to sense, decide, and strike faster than US forces can perceive.

This article has shown that the solution lies not in materiel alone but in the intellectual architecture of targeting and the doctrinal systems that govern it. By concentrating human judgment in the decide phase and delegating machine-speed functions to bounded autonomous operations, the Army can employ LAWS in accordance with the LOAC. Doctrinal reforms such as HDM3, the integration of autonomy into FM 3-60 and FM 3-84, and the recognition of autonomous systems as doctrinal means provide the conceptual framework for this transition. Organizational innovations, such as ATCs and training pipelines that include the ARAS Course, ensure that supervisory control of autonomy is both institutionalized and scalable.

The Army’s advantage will not rest on faster machines but on whether its doctrine allows human judgment to act before machine speed renders it irrelevant. By modernizing its doctrine, organizations, and training now, the Army can ensure that human orders, not adversary algorithms, retain the initiative.

 

---

Notes

1. Carl von Clausewitz, On War, ed. and trans. Michael Howard and Peter Paret (Princeton University Press, 1984), 89.
2. Koichiro Takagi, “Is the PLA Overestimating the Potential of Artificial Intelligence?,” Joint Force Quarterly 116 (2025): 71, https://digitalcommons.ndu.edu/joint-force-quarterly/vol116/iss4/10.
3. Ingvild Bode et al., “Prospects for the Global Governance of Autonomous Weapons: Comparing Chinese, Russian, and US Practices,” Ethics and Information Technology 25, no. 1 (2023): 2, doi.org/10.1007/s10676-023-09678-x.
4. Xie Kai et al., “Perspective on the Transmutation of the Winning Mechanism of Intelligent Warfare,” ed. Yang Fanfan, China Military Online—People’s Liberation Army Daily, 26 April 2022, http://www.81.cn/bz_208549/10150428.html.
5. Oscar Jonsson, The Russian Understanding of War: Blurring the Lines Between War and Peace (Georgetown University Press, 2019), 107.
6. Department of Defense (DOD) Directive 3000.09, Autonomy in Weapon Systems (DOD, 2023), 10, https://www.esd.whs.mil/portals/54/documents/dd/issuances/dodd/300009p.pdf.
7. DOD Directive 3000.09, Autonomy in Weapon Systems, 3, 21.
8. DOD Directive 3000.09, Autonomy in Weapon Systems, 1.
9. DOD Directive 3000.09, Autonomy in Weapon Systems, 3.
10. DOD Directive 3000.09, Autonomy in Weapon Systems, 15.
11. DOD Directive 3000.09, Autonomy in Weapon Systems, 10.
12. Office of the General Counsel, Department of Defense Law of War Manual (DOD, 2023), 251–53, https://media.defense.gov/2023/Jul/31/2003271432/-1/-1/0/DOD-LAW-OF-WAR-MANUAL-JUNE-2015-UPDATED-JULY%202023.PDF.
13. Field Manual (FM) 3-60, Army Targeting (US Government Publishing Office [GPO], 2023), 1-8, https://armypubs.army.mil/epubs/DR_pubs/DR_a/ARN39048-FM_3-60-000-WEB-1.pdf.
14. Elsa Kania, “Minds at War: China’s Pursuit of Military Advantage Through Cognitive Science and Biotechnology,” PRISM 8, no. 3 (2020): 86, https://ndupress.ndu.edu/Portals/68/Documents/prism/prism_8-3/prism_8-3_Kania_82-101.pdf.
15. Aleksandr A. Svechin, Strategy, ed. Kent D. Lee (East View Information Services, 1991), 69.
16. Georgii Isserson, The Evolution of Operational Art, trans. Bruce Menning (Combat Studies Institute Press, 2013), 57–58.
17. Edward Atkeson, Soviet Theater Forces at the Crossroads, Land Warfare Papers No. 1 (Institute of Land Warfare, Association of the United States Army, 1989), 17, https://www.ausa.org/sites/default/files/LWP-1-Soviet-Theater-Forces-at-the-Crossroads.pdf.
18. Jacob Kipp, “The Russian Military and the Revolution in Military Affairs: A Case of the Oracle of Delphi or Cassandra?,” paper presented at the Military Operations Research Society [MORS] Conference, Annapolis, MD, 6–8 June 1995.
19. Atkeson, Soviet Theater Forces at the Crossroads, 19.
20. Valery Gerasimov, “The Value of Science Is in the Foresight: New Challenges Demand Rethinking the Forms and Methods of Carrying out Combat Operations,” Military Review 96, no. 1 (January-February 2016): 24, https://www.armyupress.army.mil/Portals/7/military-review/Archives/English/MilitaryReview_20160228_art008.pdf.
21. Gerasimov, “The Value of Science Is in the Foresight,” 26.
22. Jack Watling and Nick Reynolds, Meatgrinder: Russian Tactics in the Second Year of Its Invasion of Ukraine (Royal United Services Institute, 2023), 13, https://static.rusi.org/403-SR-Russian-Tactics-web-final.pdf.
23. Harry Halem, “Ukraine’s Lessons for Future Combat: Unmanned Aerial Systems and Deep Strike,” Parameters 53, no. 4 (2023): 29, https://press.armywarcollege.edu/parameters/vol53/iss4/4/.
24. People’s Liberation Army (PLA) Academy of Military Sciences, In Their Own Words: Foreign Military Thought, Science of Military Strategy, 2013, trans. China Aerospace Studies Institute (Air University Press, 2022), 186–88, https://www.airuniversity.af.edu/CASI/Display/Article/2485204/plas-science-of-military-strategy-2013/.
25. Timothy L. Thomas, China Military Strategy: Basic Concepts and Examples of Its Use (US Army Foreign Military Studies Office, 2014), 64–79, https://g2webcontent.z2.web.core.usgovcloudapi.net/OEE/FMSO%20Books/2014-_-China-Military-Strategy_Basic-Concepts-and-Examples-of-its-Use-_Thomas_.pdf.
26. PLA Academy of Military Sciences, In Their Own Words, 174, 338.
27. People’s Liberation Army (PLA) Academy of Military Sciences, In Their Own Words: Science of Military Strategy, 2020, trans. China Aerospace Studies Institute (Air University Press, 2022), 29, https://www.airuniversity.af.edu/Portals/10/CASI/documents/Translations/2022-01-26%202020%20Science%20of%20Military%20Strategy.pdf.
28. PLA Academy of Military Sciences, In Their Own Words, 119.
29. Takagi, “Is the PLA Overestimating the Potential of Artificial Intelligence?,” 73.
30. State Council of China, “A New Generation Artificial Intelligence Development Plan,” trans. DigiChina, Stanford University, 20 July 2017, https://digichina.stanford.edu/work/full-translation-chinas-new-generation-artificial-intelligence-development-plan-2017/.
31. PLA Academy of Military Sciences, In Their Own Words, (2013), 317.
32. C. Anthony Pfaff and Christopher John Hickey, Integrating Artificial Intelligence and Machine Learning Technologies into Common Operating Picture and Course of Action Development (US Army War College Press, 2025), 66, https://media.defense.gov/2025/Jul/15/2003754354/-1/-1/0/20250715_PFAFF-HICKEY_AIANDML_ONLINE.PDF.
33. “How NGC2 Is Expanding the Battlefield Network at Ivy Sting 2,” Anduril Industries, 19 November 2025, https://www.anduril.com/news/how-ngc2-is-expanding-the-battlefield-network-at-ivy-sting-2.
34. Analise Callaghan, “Striking First with Machine Learning: Field Research in the 101st Airborne Division,” Modern War Journal 1, no. 1 (2025): 16–22, https://mwi.westpoint.edu/the-class-of-1982-modern-war-journal-issue-1-leadership-and-the-emerging-battlefield/.
35. Army Doctrine Publication (ADP) 1-01, Doctrine Primer (US GPO), 1-1, https://armypubs.army.mil/epubs/DR_pubs/DR_a/pdf/web/ARN18138_ADP%201-01%20FINAL%20WEB.pdf.
36. ADP 3-13, Information (US GPO, 2023), 2-14, armypubs.army.mil/epubs/DR_pubs/DR_a/ARN39736-ADP_3-13-000-WEB-1.pdf.
37. FM 3-60, Army Targeting, 2-1.
38. FM 5-0, Planning and Orders Production (US GPO, 2024), 341–43, https://armypubs.army.mil/epubs/DR_pubs/DR_a/ARN44590-FM_5-0-001-WEB-3.pdf.
39. DOD Directive 3000.09, Autonomy in Weapon Systems, 10.
40. Office of the General Counsel, Law of War Manual, 256–67.
41. FM 3-60, Army Targeting, 2-12.
42. FM 5-0, Planning and Orders Production, 160.
43. DOD, Responsible Artificial Intelligence Strategy and Implementation Pathway (DOD, 2022), 5–7, https://media.defense.gov/2024/Oct/26/2003571790/-1/-1/0/2024-06-RAI-STRATEGY-IMPLEMENTATION-PATHWAY.PDF.
44. DOD Directive 3000.09, Autonomy in Weapon Systems, 3–4.
45. FM 3-84, Legal Support to Operations (US GPO, 2023) 2-18, armypubs.army.mil/epubs/DR_pubs/DR_a/ARN39171-FM_3-84-000-WEB-1.pdf.
46. Office of the General Counsel, Law of War Manual, 203–4.
47. Annemarie Vazquez, “Laws and Lawyers: Lethal Autonomous Weapons,” Military Law Review 228, no. 1 (2020): 131, https://tile.loc.gov/storage-services/service/ll/llmlp/58062115_228-issue1-2020/58062115_228-issue1-2020.pdf.
48. Vazquez, “Laws and Lawyers,” 94.
49. Eric Jensen, “Autonomy and Precautions in the Law of Armed Conflict,” International Law Studies 96 (2020): 584, https://digital-commons.usnwc.edu/cgi/viewcontent.cgi?article=2933&context=ils.
50. Naveen Krishnan, “AI Agents: Evolution, Architecture, and Real-World Applications,” preprint, arXiv, 16 March 2025, 13, https://doi.org/10.48550/arXiv.2503.12687.
51. Surya Gangadhar Patchipala, “Tackling Data and Model Drift In AI: Strategies for Maintaining Accuracy During ML Model Inference,” International Journal of Science and Research Archive 10, no. 2 (2023): 1203, https://doi.org/10.30574/ijsra.2023.10.2.0855.
52. Krishnan, “AI Agents,” 13.
53. Kelley M. Sayler, “Defense Primer: US Policy on Lethal Autonomous Weapon Systems,” CRS In Focus No. IF11150 (Congressional Research Service [CRS], updated 26 March 2026), 1–2, https://www.congress.gov/crs-product/IF11150.
54. Margarita Konaev et al., US Military Investments in Autonomy and AI: A Strategic Assessment (Center for Security and Emerging Technology, October 2020), 19, https://cset.georgetown.edu/wp-content/uploads/U.S.-Military-Investments-in-Autonomy-and-AI_Strategic-Assessment-1.pdf.
55. FM 3-60, Army Targeting, 2-11.
56. Patchipala, “Tackling Data and Model Drift in AI,” 1199–1203.
57. DOD Directive 3000.09, Autonomy in Weapon Systems, 2, 8–10.
58. FM 3-84, Legal Support to Operations, 3-17.
59. FM 3-84, Legal Support to Operations, 2-19.

 

 

Back to Top