Abstract: Real-world adversaries constantly change and adapt. AI can generate such realistic opponents for real-time wargaming scenarios. AI-augmented training can enhance human analytical capabilities rather than replacing them. This allows young officers to be exposed to more realistic scenarios during training, resulting in more confident decision-making, adaptability, and creativity.
Problem statement: How can AI-generated adversaries be used in red teaming to improve tactical and operational decision-making in officer training?
So what?: Military training institutions and operational commands should consider integrating AI-assisted red teaming into both formal training and field exercises. This requires a clear doctrinal framework, instructor training, and ongoing evaluation to ensure that AI serves as a genuine educational tool rather than a technological gimmick. The focus should not be on the complexity of the system used, but rather on the quality of the decision-making processes it helps to develop.

The Case for Adaptive Officer Training
“Lessons are not automatically learned. They are learned only when you develop new concepts, change doctrine, and revisit training.”
General David H. Petraeus (U.S. Army, Ret.)[1]
The contemporary security environment is characterised by increasingly adaptive adversaries operating across military, informational, cyber, and economic domains. NATO defines hybrid threats as the coordinated use of military and non-military, covert and overt means designed to blur the distinction between war and peace.[2] Similarly, the European Union identifies sabotage, foreign information manipulation and interference (FIMI), malicious cyber activities, and the instrumentalisation of migration as key components of modern hybrid campaigns.[3] For military leaders, this means confronting opponents who continuously adapt their methods, exploit ambiguity, and seek to create uncertainty in the decision-making environment.
Recent events underscore this reality. In February 2026, Russia launched one of its most significant attacks on Ukraine’s energy infrastructure since the start of the full-scale invasion, demonstrating once again how contemporary adversaries combine military action with broader efforts to disrupt critical systems and undermine societal resilience.[4] Such developments reinforce a fundamental reality of modern conflict: military leaders must make decisions under conditions of uncertainty against opponents who learn, adapt, and react in real time.
Yet many officer training environments remain comparatively structured and predictable. Training exercises are often designed around predefined learning objectives, scripted adversary behaviour, and tightly controlled scenarios. While these approaches are essential for teaching doctrine and foundational competencies, they can inadvertently reward pattern recognition rather than adaptability. When officers become familiar with an exerciseis design, they may learn how to succeed within the scenario rather than how to respond to an intelligent, adaptive opponent. Military practitioners have long warned against this tendency. As one observer noted following a major exercise, adversaries should never be “waved off”; they are “free-thinking” opponents whose purpose is to surprise and challenge assumptions.[5] Effective wargaming must therefore expose participants to the same uncertainty and adversarial friction they are likely to encounter in real operations.[6], [7]
Red teaming has emerged as one of the most valuable tools for achieving this objective. The British Ministry of Defence’s Red Teaming Handbook emphasises that red teaming is not simply about role-playing an adversary but about challenging assumptions, exposing blind spots, and improving the quality of decision-making.[8] However, effective red teaming is resource-intensive. Creating realistic and adaptive opposing forces requires experienced personnel, extensive preparation, and continuous instructor involvement. As a result, training institutions often face a trade-off between realism, scalability, and educational control.
Artificial intelligence (AI) offers a potential solution to this challenge. Not as a replacement for instructors or military judgement, but as a means of generating adaptive adversarial behaviour, increasing scenario variability, and exposing officers to a broader range of plausible operational challenges. Organisations facing rapid technological change, including the U.S. Air Force, have already recognised the need to adapt doctrine and training “at the speed of relevancy.”[9] The question is therefore not whether officer education should continue to develop adaptability and critical thinking, but how training systems can provide more realistic and responsive adversaries at scale.
The Factor of Uncertainty
The central challenge of officer training in this context lies not in whether officers understand doctrine, but in whether they can make sound decisions even when doctrine alone does not provide a clear answer. Contemporary conflicts are characterised by ambiguity, combat in the information domain, and adaptive adversaries who are constantly reshaping the operational environment. The training of military leaders should therefore not merely aim to apply established procedures but also prepare them to adapt when these procedures reach their limits. It is therefore necessary to examine more closely the role of uncertainty in military decision-making processes and the development of leadership competence before red teaming can be explored as a potential solution.
Military decisions should not centre around identifying a single optimal solution. Instead, they should be about making timely, effective decisions under pressure and with incomplete information, as in combat, the situation changes rapidly and waiting for complete information can be fatal.[10] Officers may operate under bounded rationality, with limited time and intelligence, and amid high stakes. Therefore, the true aim of officer training should be to improve recognition, adaptation, and resilience in the face of the unexpected, rather than to provide perfect answers.[11]
Standardised training nevertheless remains an indispensable foundation of officer education. It familiarises officers with doctrine, tactics, techniques, and procedures; establishes a common technical language; and imparts the basic competencies on which military effectiveness depends. The argument, therefore, is not that standardisation should be abandoned, but that it must be purposefully supplemented. Once fundamental knowledge and procedural skills are firmly established, officers must learn to act even under conditions where doctrine offers no clear answer, the situation changes faster than any plan can anticipate, and adversaries deliberately exploit rigidity. It is precisely at this transition from procedural certainty to adaptive judgment that training that deliberately works with uncertainty and promotes critical reflection demonstrates its greatest value.
Training scenarios should seek to develop an officer’s ability to recognise patterns by drawing on experience quickly, adapting to new developments, and recovering when things do not go according to plan. Adversaries react to moves in real time. Even the best-laid plans can (and most likely will) degrade upon first contact with the enemy, a phenomenon famously summarised as “no plan of operations extends with any certainty beyond the first contact with the main hostile force”.[12] Furthermore, in a phenomenon known as the ‘fog of war,’ information in combat often arrives late, comes in fragments, or proves misleading.[13] Officer training acknowledges this phenomenon by seeking to inoculate leaders against shock and confusion. Rather than teaching a rigid formula, good training should put officers in situations where they must interpret ambiguous information amid the ‘fog of war’, make decisions with risk, and adjust on the fly as the situation evolves.
Crucially, exposure to uncertainty must be deliberate and intentionally integrated into the training design. If an exercise is purely random or chaotic with no learning objective, officers may become confused or frustrated. Instead, instructors must introduce structured uncertainty by forcing a decision-maker to reassess their plan at critical points. For example, a platoon leader in training might receive an unexpected report of an enemy flanking manoeuvre mid-mission, not to throw them off, but to assess how they re-evaluate their plan in light of a new threat. The value of red teaming in this context is that it adds friction by complicating the scenario in ways that test the officer’s assumptions and reactions, thereby yielding teachable moments. An effective red team does not just make the scenario harder for difficulty’s sake; it should present a plausible surprise that forces the officers to think critically and adapt.
Beyond building adaptability, training in uncertain conditions must also prepare officers for failure. When things go wrong, as they inevitably will in real combat, officers must be able to quickly assess a deteriorating situation and find a new course of action. Challenging, realistic training is crucial for precisely this reason: it develops not only tactical knowledge but also cognitive agility, the ability to think clearly and lead effectively when real chaos breaks out. [14], [15]
Red Teaming as a Mindset
When uncertainty is accepted as a defining feature of modern conflict, military education requires methods that expose officers to it in a structured, educationally useful way. Red teaming represents one such approach.
Red teaming is fundamentally about challenging the way we think, rather than simply role-playing an enemy force. In military contexts, the purpose of a red team is to surface hidden assumptions, expose blind spots, and test unexamined beliefs in plans or strategies.[16] A well-run red team should compel planners and decision-makers to justify their choices and consider alternative interpretations of the same situation. The goal is not to ‘win’ as the enemy but to provoke reflection. In a red teaming exercise, the red team challenges the main team’s assumptions, strategies, and blind spots, not to criticise, but to illuminate what the team might have missed due to overconfidence or groupthink.[17] This process surfaces alternative ways of thinking and encourages critical examination of a plan from multiple angles.
The educational value of red teaming is thus well established. The more difficult question, however, is why red teaming often falls short of its potential in officer training, despite its recognised benefits. Common practice in many exercises reduces red teams to scripted opposition forces. These red teams are often constrained by safety rules, tight schedules, and limited instructor manpower. Over time, students can anticipate the red team’s behaviour, defeating the purpose. As a result, the ‘enemy’ in training often can then follow a known script. “When red teams become predictable, they no longer challenge judgement; they validate pattern recognition.”[18] In other words, if the adversary always does exactly what the scenario designers expect, officers are not truly being challenged to think; they are just rehearsing a response.
Importantly, this critique is not aimed at blaming institutions or instructors; it highlights structural limits. Safety considerations rightly prevent true ‘free play’ by opposing forces (e.g. exercises cannot allow uncontrolled escalation or unsafe situations). Time and workload constraints mean instructors often reuse scenarios and known adversary behaviours. This resulting predictability is a problem. If every class of officers faces the same scripted scenarios, they may learn to beat the exercise rather than adapt and respond to the unexpected. As one expert noted in hindsight, “with fewer institutional challenges, planners focus on an adversary’s actions but neglect their own assumptions. This produces plans built on assumptions that have not been fully stress-tested”.[19] In short, a red team can inadvertently reinforce students’ existing mental models rather than breaking them. If red teaming is about thinking against oneself, its effectiveness depends on exploring many alternative paths. Yet, due to the limits, human-led training exercises can only cover a few possibilities.
The Limits of Traditional Training Red Teams
Despite their educational value, human-led red teams face significant practical constraints that can gradually reduce their effectiveness in training environments. One major constraint is the capacity of instructors and support staff. Crafting and running a dynamic adversary in an exercise is labour-intensive. There are only so many tricks a small cadre of instructors can pull, and safety or resource limitations often force them to stick to a certain script. Exercises are often reused across multiple classes or rotations, which means adversary behaviours become repetitive and well-known. Over the years, an “exercise lore” can develop as officer cadets hear stories from predecessors about how the red team behaved, and they come in ready for those known patterns instead of facing a truly new challenge.
Another limitation is that red team behaviour is often shaped by the training objectives and schedule. In a typical field exercise, the opposing force might be instructed to put up a good fight but ultimately allow the trainees to achieve the lesson objective (for example, successfully defending a hill or completing a convoy). Known escalation thresholds mean the red team will only attack up to a certain intensity, with scenario injects scripted at predetermined times. These boundaries exist for valid reasons (safety, ensuring specific learning points are hit, time management), but they also make the red team’s behaviour predictable. Officer students quickly learn enemy moves by heart. This problem is further reinforced by the fact that performance evaluations and incentive structures often reward visible success rather than appropriate decision-making in the situation. This can encourage officers to focus on delivering what they know evaluators want to see to secure a good grade, rather than engaging with the scenario as a genuine learning experience.
Over time, officerstudents may adapt less to true uncertainty and more to the exercise itself. If every year the red team ambushes the convoy at the same valley, then by year three, the new officers are already posting extra scouts on the hill because they know an ambush is coming (having heard from last year’s class). The training then ceases to test adaptability. It instead validates that the students learned about the exercise. One military commentator observed that when institutional red teaming atrophied, planners became overly focused on expected enemy moves and “neglect their own assumptions,” leading to plans that succeed in scripted wargames but would falter in reality. Similarly, in a training context, if the ‘enemy’ always behaves in a textbook way, students are rewarded for pattern-matching rather than for true problem-solving. The consequence of these limits is a reduced element of surprise and a narrower decision space for learners. If nothing unexpected ever happens, then officers have little incentive to question their initial assumptions. This is antithetical to the spirit of red teaming, which should constantly challenge and unsettle plans.
It is important to note that the challenge is not realism for its own sake. The real challenge is achieving responsive opposition that dynamically reacts to the officer’s decisions. In real operations, if a platoon does something against the enemy, the enemy does not keep following a script; they adapt back. Traditional exercises struggle to emulate that adaptiveness. Once a scenario is written, it tends to play out one way. If students take an inventive approach, a human red cell might not be empowered to deviate from the script to counter it. Thus, officer students do not get to see the second- and third-order consequences of their actions. They see only the scripted chain. The real constraint, then, is operational: producing an adversary that remains adaptive, varied, and responsive across repeated training cycles and large cohorts of officers is simply very hard to do at scale.
How AI Adds Value
AI systems capable of generating and adapting adversarial behaviour offer the possibility of a more dynamic and scalable red-team opponent. Whether employed through large language models that generate adversary responses and scenario injects, or through simulation systems that adapt to trainee actions, these approaches create greater variability and responsiveness than scripted training environments typically allow. Their role, however, must remain that of a support tool for training rather than a replacement for human judgement. Properly used, AI can serve as a “sparring partner generator,” creating a wide range of plausible adversary behaviours and responses to push officers in training. In short, AI should be used togenerate variation, augment human-led red teaming, andscale up the options that a red team can throw at students without taking away the instructors’ authority to guide the exercise.
Expanded Adversary Variability
Traditional simulations often rely on pre-scripted enemy courses of action, which can create repetition and reduce the realism of training over time. AI-generated adversaries, by contrast, can produce multiple plausible responses to the same Blue Force action. Rather than executing a fixed sequence, an AI system can select or generate tactics dynamically based on how the scenario unfolds.[20]
Wargaming research emphasises that the value of simulations lies not merely in rehearsing procedures, but in exposing participants to the consequences of different decisions and the alternative pathways those decisions generate. Comparative wargaming seeks to explore how varying conditions influence decision-making and outcomes.[21] AI-generated adversaries may expand this logic by increasing the number and diversity of plausible adversary responses that can be explored within a single training environment. The result is exposure to a broader range of operational contingencies and “what-if” outcomes. By reducing predictability, AI-augmented red teaming better reflects the adaptive behaviour of real-world adversaries and trains officers to expect uncertainty rather than scripted progression.
Responsive Scenario Evolution
AI-enabled red teams can move beyond fixed injects and linear exercise design. Instead of relying on predetermined triggers, an AI adversary can adapt in real time to trainee decisions. If one axis is secured, pressure may shift elsewhere; if a vulnerability emerges, it can be exploited immediately.
This move–countermove dynamic produces non-linear scenarios in which Blue and Red actions continuously influence one another. The exercise becomes an evolving interaction rather than a sequence of staged events. Such adaptive scenario development reinforces agility and decision-making under fluid conditions, mirroring the behaviour of a thinking opponent rather than a static script.
This responsiveness also enables officers to observe the second- and third-order consequences of their decisions. A tactical success may generate new vulnerabilities elsewhere, while a choice that appears advantageous in the short term may create operational, informational, or logistical challenges later in the scenario. Officers must therefore continuously reassess how their actions shape the evolving environment, rather than simply executing the next step in a predetermined sequence.
Cognitive and Informational Domains
The value of AI-augmented red teaming may be particularly significant in cognitive and informational domains, where adversaries increasingly compete through influence, deception, and information manipulation rather than solely through kinetic action. Since AI red teaming is not limited to kinetic operations, it can be particularly valuable in cognitively demanding domains such as information warfare, deception, and negotiation. Simulating disinformation campaigns, adversary messaging, or complex civilian interactions traditionally requires extensive instructor bandwidth and consistent role-play execution. AI agents can generate and adapt narratives at scale, simulate adversary personas, and test officers’ responses to misinformation or diplomatic engagement in real time. This enables systematic training in cognitive and informational environments that are operationally decisive yet difficult to rehearse repeatedly with limited personnel. By embedding adaptive narrative and psychological elements into exercises, AI broadens red teaming beyond battlefield manoeuvre into the full spectrum of modern conflict. At the same time, outputs generated in these domains require careful oversight, as AI systems may reproduce assumptions embedded in their training data rather than accurately reflecting how specific adversaries communicate, persuade, or deceive.
Clear boundaries are essential for the responsible use of AI in officer training. Rather than replacing instructors or officer judgement, AI should function as a tool that generates adversarial behaviour, scenario variation, and alternative courses of action. This distinction matters: AI’s value lies not in making decisions, but in creating the conditions under which officers must make them.[22] Scholars have cautioned that AI should enhance, rather than replace, human expertise, and that the judgement of the officer corps must remain central to any augmented training environment.[23], [24] Keeping AI in a supporting role, as a scenario engine that challenges assumptions and expands the range of decision-making situations officers encounter, should therefore be both a pedagogical and an institutional imperative.
Risks, Failure Modes, and Why Scepticism is Essential
Before rushing to put AI at the helm of training adversaries, the risks and potential failure modes must be addressed. Any new technology introduced into military training carries epistemic, cognitive, and institutional risks that, if unaddressed, may undermine the very learning outcomes it is intended to enhance.
Epistemic Risks: Knowledge, Logic, and Model Integrity
AI-generated adversaries may appear plausible while resting on flawed assumptions or embedded biases. An AI red cell is only as reliable as its underlying models and training data. If those inputs contain distortions, the system may produce behaviour that looks realistic yet diverges from how a real adversary would act.[25] For example, an AI might systematically overestimate an opponent’s willingness to accept casualties, leading it to adopt aggressive or suicidal tactics that a rational adversary would avoid. If such patterns go unexamined, trainees risk internalising inaccurate assumptions about adversary doctrine and decision-making.
Scholars have warned that AI can encode biases in training data, obscure reasoning processes through algorithmic opacity, and generate confident outputs based on unvalidated assumptions.[26] In a training environment, this epistemic opacity is particularly dangerous: realistic outputs may mask flawed logic. Mitigation requires rigorous validation and oversight. AI-generated tactics should be reviewed by subject-matter experts and assessed against real adversary doctrine. AI outputs must remain subject to critical scrutiny rather than be treated as authoritative simply because they are machine-generated.
Learning, Judgement, and Human Agency
AI integration also introduces cognitive risks that affect how officers learn and make decisions. One central concern is automation bias, the tendency to defer to machine-generated outputs without sufficient critical evaluation.[27] The speed and confidence of AI systems can encourage undue trust, potentially discouraging independent judgement. In a training context, this would undermine the very objective of red teaming: developing critical thinking under uncertainty. If officer students begin to rely on AI-driven elements rather than interrogating them, the exercise’s educational function weakens.
A second cognitive risk is complexity. AI systems capable of generating highly adaptive responses may produce overly intricate scenarios that obscure the intended learning objectives. Training differs from war in that it is deliberately structured to teach specific lessons. An AI that continuously optimises against the trainee risks overwhelming rather than educating.
There is also the risk of reduced ownership. If officers perceive scenarios as primarily driven by opaque algorithms rather than guided by instructors, psychological engagement and accountability may diminish. Maintaining visible human oversight is therefore essential to preserve trust, clarity of purpose, and pedagogical control.
Organisational Incentives and Governance
At the institutional level, the introduction of AI may subtly reshape training priorities.[28] The availability of advanced simulation tools can create pressure to maximise usage rather than maximise learning. An institution may equate a higher number of AI-generated scenarios with improved preparedness, even when pedagogical depth is sacrificed. This “tool-driven drift” risks shifting focus from foundational competencies to technological novelty. Without deliberate integration into curriculum design, AI can produce expansion without coherence.
Security and data governance present additional concerns. AI systems often require substantial amounts of data, potentially including sensitive operational or doctrinal information. Questions of storage, access, model training, and vulnerability to compromise must be addressed rigorously. The use of commercial platforms further increases the risk of unintended exposure to training concepts or institutional biases.
Finally, cultural dynamics matter. If AI adoption is driven by institutional enthusiasm rather than educator buy-in, implementation may become performative rather than purposeful. The central objective must remain learning outcomes rather than technological innovation. The greatest danger lies not in misuse alone, but in uncritical adoption detached from pedagogical necessity.
Having laid out these risks, it is equally important to discuss safeguards for responsible AI Red Teaming:
- Human-in-the-Loop
AI-enabled red teams must remain under positive human control. Exercise directors should retain the authority to pause, modify, or override AI-driven developments at any point. This ensures alignment with training objectives and prevents drift into unintended scenario paths.
Dedicated monitoring roles, such as a white cell member supervising AI outputs, can provide real-time oversight. AI literacy among instructors is essential to ensure that human judgement remains decisive rather than symbolic.
- Bounded Scenario Design
AI behaviour must operate within clearly defined parameters that reflect real-world doctrine and exercise constraints. Scenario boundaries, such as rules of engagement, adversary capabilities, and escalation limits, should be explicitly encoded to prevent unrealistic or doctrinally unsound outcomes. Constraining the system within a defined sandbox preserves realism while preventing uncontrolled scenario escalation.
- Instructor-Led Adjudication and Meaning-Making
AI may generate events, but instructors remain responsible for adjudication and interpretation. Mechanisms for intervention, pausing, reverting, and simplifying must be readily available. Post-exercise analysis should remain instructor-led. AI can produce data; human educators translate events into doctrine, judgement, and learning.
- Continuous Validation of the AI System
AI adversaries should themselves be subject to rigorous testing. Before operational use, subject-matter experts should stress-test behaviour across diverse scenarios to detect bias, logical flaws, or unrealistic patterns. Validation should be continuous rather than a one-time certification. Where possible, explainability features should allow instructors to understand the reasoning behind AI decisions.
- Student Education and Cognitive Framing
Officers must be trained not only to defend against AI-driven adversaries but also to engage critically with AI-generated outputs. Students should neither treat AI behaviour as infallible nor dismiss it as artificial. Transparency about AI involvement can reinforce critical evaluation, prompting officers to question adversary actions and connect them to doctrinal reasoning.
In sum, a careful approach can address these pitfalls. The introduction of AI into training must be accompanied by strict ethical and practical guidelines; always as a tool, never the ultimate arbiter, with human supervision, and constantly evaluated for unintended effects. When anchored in these safeguards, an AI-supported red teaming program can be developed to combine technological capability with human pedagogical oversight.
Responsible AI Red Teaming
Looking ahead, any implementation of AI-augmented red teaming should adhere to a few key design principles to ensure it enhances training instead of derailing it.
The first principle is trivial, but often crucial: first the learning objective, then the tool. Game and exercise design must prevent method and technology from dominating the issue at hand. Not only is it essential to identify the goal of the game first so that the impact can later be measured, but it can also be helpful to deliberately “scope”[29] games to ensure the design meets its objective. More precisely, AI should primarily help to generate opposing behaviour and surprise, not lead the exercise.
The key point is therefore that humans remain in control of decision-making and evaluation. Especially when adjudication is not completely rigidly rule-bound, there is pressure for transparency. Equally important: positive control by the exercise management is needed to ensure that the game remains within the parameters of the question.[30] Therefore, part of the decision-making authority must lie with the game staff to ensure the AI acts within clear guardrails. At Hadean, they put that into a practical model with the principle of “Immediately Validatable Output (IVO)”:[31] AI should only generate content that humans can immediately verify and check for plausibility against doctrine/tables.
A responsible approach also makes targeted use of AI to reveal ramifications and second- and third-order effects, without allowing the exercise to descend into arbitrary complexity. Elizabeth Bartels describes the design goal for “alternative conditions” games as minimising variation beyond the factors of interest to enable pattern identification from comparative runs.[32] This logic can be transferred didactically: AI can generate alternative Red COAs (branching), while instructors control which variables are deliberately varied and which must remain constant.
Ultimately, instructor curation and the interpretation of outcomes remain central. The Dupuy case study shows how much learning depends on instructors providing advice on decisions, adjudicating results in an understandable way, and then critiquing/evaluating them.[33] At the same time, acceptance of the instructor and method is a didactic factor. The use of AI must therefore also be motivational: understandable outputs, comprehensible logic, and an exercise dramaturgy that supports learning.
Technologies Change, Leadership Endures
The incorporation of AI into red teaming is a means to an end, not an end in itself. The real challenge in officer training lies not in a lack of doctrine, but in preparing officers to exercise effective judgment under conditions of uncertainty and against adaptive adversaries who do not follow scripts, do not respect boundaries, and do not behave as expected. AI-assisted red teaming offers a way to increase exposure to precisely these conditions: the variability, friction, and adversarial responsiveness that traditional, script-based training environments are structurally unable to sustain to the necessary extent.
However, its effectiveness depends entirely on how it is implemented. An AI system that generates realistic adversarial behaviour but is operated without instructor oversight, without a doctrinal foundation, or without continuous validation does not solve the problem. It replicates it in a more sophisticated form. AI in training is, at its core, a design task, not a shortcut. It requires the same rigour, iterative refinement, and institutional commitment that good doctrine has always demanded. Technologies evolve and are replaced, whereas adaptable leadership endures.
The measure of success, therefore, is not the complexity of the AI system, but whether it produces officers who are better able to question assumptions, adapt under pressure, and lead effectively when plans collide with reality. General Petraeus reminds us that institutions do not learn on their own. They learn when they are willing to rethink doctrine and adapt their training methods. This willingness, not the technology itself, will determine whether AI-assisted red teaming delivers on its promise.
[1] Jillian Kay Melchior, “NATO Has Seen the Future and Is Unprepared: A Simulation of Drone Warfare Shows How Far the Alliance Has to Go to Learn the Lessons of Ukraine,” Wall Street Journal, February 12, 2026, https://www.wsj.com/opinion/nato-has-seen-the-future-and-is-unprepared-887eaf0f.
[2] NATO, “Countering Hybrid Threats,” accessed January 29, 2026, https://www.nato.int/en/what-we-do/deterrence-and-defence/countering-hybrid-threats.
[3] Council of the European Union, “Hybrid Threats,” accessed January 29, 2026, https://www.consilium.europa.eu/en/policies/hybrid-threats/.
[4] David L. Stern and Serhii Korolchuk, “Russia Delivers Worst Attack This Year to Ukraine’s Power Sector,” Washington Post, February 3, 2026, https://www.washingtonpost.com/world/2026/02/03/trump-war-russia-ukraine-attacks-power-sector/.
[5] Melchior, “NATO Has Seen the Future.”
[6] Center for Army Lessons Learned, How to Master Wargaming: Commander and Staff Guide to Improving Course of Action Analysis, no. 20-06 (Fort Leavenworth, KS: Center for Army Lessons Learned, 2020).
[7] Ibid.
[8] UK Ministry of Defence, Red Teaming Handbook (London: Ministry of Defence, 2021), https://assets.publishing.service.gov.uk/media/61702155e90e07197867eb93/20210625-Red_Teaming_Handbook.pdf.
[9] U.S. Air Force, Air Force Doctrine Publication 3-5: Information, accessed January 29, 2026, https://www.doctrine.af.mil/home/AFD35/.
[10] Center for Army Lessons Learned, “Military Decision-Making Process: Organizing and Conducting Planning” (United States Army, 2023), https://api.army.mil/e2/c/downloads/2023/11/17/f7177a3c/23-07-594-military-decision-making-process-nov-23-public.pdf.
[11] Jacob Hawkins, “Educating, Training, and Inspiring Officers to Lead Adaptively with Character in a Complex 21st Century Environment,” M-RCBG Associate Working Paper Series No. 138 (Cambridge, MA: Harvard Kennedy School, 2020), https://www.hks.harvard.edu/sites/default/files/centers/mrcbg/files/Final_AWP_138_1.pdf.
[12] Helmuth von Moltke, Moltke on the Art of War: Selected Writings, ed. Daniel J. Hughes and Harry Bell (Novato, CA: Presidio Press, 1993), https://archive.org/details/moltkeonartofwar0000molt/page/n9/mode/2up.
[13] Norwich University, “21st Century Fog of War: Information Warfare and the Battle for Perception,” October 14, 2024, https://www.norwich.edu/news/21st-century-fog-war-information-warfare-and-battle-perception.
[14] Jeff DeGraff, “The Age of Answers Is Over: Training Adaptive Thinkers for Modern Military Realities,” Modern War Institute, May 23, 2025, https://mwi.westpoint.edu/the-age-of-answers-is-over-training-adaptive-thinkers-for-modern-military-realities/.
[15] Command Sgt. Maj. LeVares J. Jackson Sr., “Tough, Realistic Training Is a Form of Taking Care of Soldiers,” NCO Journal, February 2025, https://www.armyupress.army.mil/portals/7/nco-journal/images/Muddy-Boots/Tough-Realistic-Training/Tough-Realistic-Training.pdf.
[16] Subho Majumdar, Brian Pendleton, and Abhishek Gupta, “Red Teaming Is a Critical Thinking Exercise: Part 1,” AVID Blog, April 16, 2025, https://avidml.org/blog/red-teaming-1/.
[17] Australian Defence Force, “Forgotten by the ADF: Red Teaming and Alternative Analysis,” Future Forge, 2021, https://theforge.defence.gov.au/war-college-papers-2023/forgotten-adf-red-teaming-and-alternative-analysis.
[18] Semi Yulianto, Benfano Soewito, Ford Lumban Gaol, and Aditya Kurniawan, “Enhancing Cybersecurity Resilience through Advanced Red Teaming Exercises and MITRE ATT&CK Framework Integration: A Paradigm Shift in Cybersecurity Assessment,” Cyber Security and Applications 3 (2024): 100077, https://doi.org/10.1016/j.csa.2024.100077.
[19] Joseph Wehmeyer, “Who Stress-Tests U.S. War Plans?,” War on the Rocks, January 8, 2026, https://warontherocks.com/2026/01/who-stress-tests-u-s-war-plans/.
[20] Matyas Ortmann, “Artificial Intelligence in Combat Simulations: How AI Is Changing NATO Soldier Training,” Defense Magazine, August 26, 2024, https://www.defensemagazine.com/article/artificial-intelligence-in-combat-simulations-how-ai-is-changing-nato-soldier-training.
[21] Erik Lin-Greenberg, “Evidence from Wargames,” in The Remote Revolution: Drones and Modern Statecraft (Ithaca, NY: Cornell University Press, 2025), 50–52.
[22] Wehmeyer, “Who Stress-Tests U.S. War Plans?”
[23] Tamir Libel, “The Impact of Artificial Intelligence on the Officer Corps,” Defence Studies 25, no. 3 (2025): 628–43, https://doi.org/10.1080/14702436.2025.2466654.
[24] Hyokwon Jung, “A Glimpse into the Future Battlefield with AI-Embedded Wargames,” Proceedings, U.S. Naval Institute, June 2024, https://www.usni.org/magazines/proceedings/2024/june/glimpse-future-battlefield-ai-embedded-wargames.
[25] Davit Khachatryan, “Military AI Challenges Human Accountability,” Center for International Policy, July 21, 2025, https://internationalpolicy.org/publications/military-ai-challenges-human-accountability/.
[26] Ibid.
[27] Drew F. Lawrence, “Military Warns EOD Techs against Uploading ‘Highly Sensitive’ Bomb Disposal Data into Generative AI Systems,” DefenseScoop, February 10, 2026, https://defensescoop.com/2026/02/10/military-warning-eod-tech-data-generative-ai-systems/.
[28] Khachatryan, “Military AI Challenges Human Accountability.”
[29] Elizabeth M. Bartels, Building Better Games for National Security Policy Analysis: Toward a Social Scientific Approach (PhD diss., Pardee RAND Graduate School, 2020), https://www.rand.org/content/dam/rand/pubs/rgs_dissertations/RGSD400/RGSD437/RAND_RGSD437.pdf.
[30] Ibid.
[31] Dan Gorringe, “King’s Wargaming Week 2025: A Case Against Faster Horses,” Hadean, June 4, 2025, https://hadean.com/blog/kings-wargaming-week-2025-a-case-against-faster-horses/.
[32] Bartels, Building Better Games.
[33] Christopher A. Lawrence, “Wargaming 101: The Use of Wargames in Training,” Dupuy Institute, August 1, 2023, https://dupuyinstitute.org/2023/08/01/wargaming-101-the-use-of-wargames-in-training/.








