Abstract: The integration of artificial intelligence (AI) into military domains, particularly in Lethal Autonomous Weapons, raises significant ethical and legal concerns in a contested and incomplete normative framework. This paper critically examines these challenges to assess the adequacy of existing legal norms and ethical frameworks governing the deployment of autonomous and semi-autonomous systems. The paper examines the implications of autonomy, accountability, and human control in lethal decision-making. Building on this analysis, the paper explores mitigation strategies, including meaningful human control, system design constraints, and life-cycle oversight.
Problem statement: How can armed forces integrate AI into military operations in a way that preserves ethical judgment and legal accountability?
So what?: States and military institutions must foster continuous dialogue to establish consensus on the definition, applications, and risks of AI use for security purposes, and to create a common, adaptive legal, governance, and ethical framework that evolves with AI technology.
Conduct and Perception of Warfare
Recent advances in military technology, particularly in artificial intelligence (AI) and unmanned systems, have profoundly transformed the conduct and perception of warfare. Designed with simplified, intuitive interfaces and increasingly based on dual-use platforms already familiar to civilian users, these technologies offer undeniable strategic advantages to armed forces. They reduce research, development, and training time and costs (as well as human costs), as military institutions can rely on existing civilian technological ecosystems, infrastructure, and experience to acquire, maintain, and operate such systems. Nevertheless, the very characteristics that render these technologies efficient also generate significant ethical and legal challenges that military leaders and policymakers must critically evaluate.[1]
An example of this is AI’s ability to overcome, for instance, the physical and moral constraints of human drone pilots. Contemporary drone warfare is often portrayed as a form of remote combat mediated through digital interfaces resembling video-game environments, causing the pilots to be detached, disengaged, and emotionally disconnected from their targets.[2] However, despite the lower risk than on the physical battlefield, they are not immune to physical strain and fatigue, emotional distress, and cognitive overload, leading to reductions in operational performance, slower reaction times, and decision-making errors.[3] The reasons for these, other than trauma from causing and witnessing loss of life in high definition, may be the blurring, or lack of compartmentalisation, between combat and personal life (in the case of American Unmanned Aerial Vehicle pilots in the Middle East, the “cockpit” may be in the U.S., not too far from the pilot’s home),[4] or what is defined as distant or temporal intimacy between operator and target.[5] AI’s immunity to physical and emotional strain, its ability to produce continuous, consistent battlefield effects, and its potential to delegate decisions and replace humans have important implications for both military ethics and international humanitarian law (IHL), blurring the boundary between virtual and actual warfare. These concerns are further exacerbated by the integration of AI into Lethal Autonomous Weapon Systems (LAWS), and the call that this improves mission effectiveness and makes warfare more humane and ethical.[6]
From a jurisprudential and ethical perspective, the most pressing question is whether it can ever be morally or legally acceptable for an autonomous system to independently decide to take human life. Ethical principles often serve as conceptual foundations for legal norms, yet many ethical considerations surrounding the use of AI-driven weapons remain beyond the scope of codified international law. This debate lies at the intersection of ethics, technology, and international law, where the normative framework remains contested and incomplete.
Ethical and Legal Challenges in the Use of Artificial Intelligence in Warfare
The increasing use of AI in military operations raises complex ethical and legal questions that echo concerns about accountability, moral agency, and the alignment of human values with machine actions.[7] Decisions that were traditionally taken entirely by humans may now involve AI systems; however, employing AI in value judgements or assessments could create an accountability gap in which legal and moral responsibility do not coincide.[8]
This is particularly the case when Agentic AI is employed, for example, in Battlefield Management Systems (BMSs). Unlike Reactive or Generative AI, Agentic AI has independent agency, the ability to model itself and plan its actions.[9] It represents multi-level autonomous, goal-setting, and goal-oriented agents capable of independent reasoning, proactive decision-making, and action execution in dynamic and complex environments.[10] All this is done at speeds exceeding human processing, with tactical and strategic systems supporting 95% and 80% automation, respectively.[11] However, a major aspect to consider, especially in the attribution of responsibility and accountability, is emergent behaviour. It is a characteristic of complex systems and their interactions and relationships among system elements to behave in unexpected, unprogrammed, or not readily explainable ways in dynamic environments. For instance, in a BMS, emergent behaviour may corrupt its goal-seeking element, leading to unreliable intelligence or inaccurate weapon-system targeting.[12]
Ethical and Moral Dimensions of AI in Warfare
New challenges related to integrating AI into military applications, particularly in autonomous weapons and systems, primarily revolve around ethical implications. More precisely, what is being questioned is the moral status of machines themselves, i.e., whether or not it is “fair” or “correct” for a LAWS to autonomously take a life. In this sense, the use of autonomous weapons undermines traditional notions of accountability, which have long served as a deterrent against harmful actions, ensured adherence to legal standards, and assigned moral responsibility for actions taken. The central argument is that autonomous weapons cannot provide adequate accountability, as it does not, in principle, seem logical to hold a non-human weapon system responsible for its actions, echoing a famous IBM training manual excerpt, “[a] computer can never be held accountable, therefore a computer must never make a management decision”.[13] This condition creates a so-called “moral buffer” that mitigates the responsibility of the military operators or commanders using the system, as well as the system’s developers, programmers, or testers.[14] Additionally, the difficulty of attribution poses an accountability challenge for AI military applications in the cyber domain, as holding perpetrators accountable for false-flag operations can be difficult.[15] As a result, advocates argue that there is an accountability gap for autonomous weapons in war, as there are no proper focal points of responsibility.
Furthermore, the inherent notion of human dignity, which is a significant ethical and religious principle, serves as the foundation of freedom, justice, and peace,[16] which would make it morally wrong for autonomous machines to take human life. This principle is enshrined, for instance, in the Universal Declaration of Human Rights, which, although not binding, has influenced international law and the development of binding human rights treaties. According to this principle, only humans, who possess said dignity and have the ability to carry out moral judgments, are “qualified” to kill another human, unlike non-human systems, which lack the necessary moral qualities to justify their actions and should not make decisions with such significant ethical implications.[17] The discussion around this issue is moreover related to the moral significance of human emotions in warfare; while emotions such as anger or fear may lead a person to act instinctively, albeit in an imprudent manner, they also provide a crucial sense of compassion and respect for human life that is necessary in exercising caution during armed conflicts; qualities that fully autonomous weapons systems do not possess. This also entails incompatibility with the moral architecture of armed conflict, as IHL assumes human judgment to be the locus of responsibility.[18]
An additional ethical risk associated with AI is its potential to undermine human rights and individual privacy. Its reliance on large databases for big-data analysis, face recognition, and cyber capabilities, as well as on networked devices that compose the Internet of Things, could enable powerful companies and extremist governments to surveil their citizens, target dissenters, and censor content.[19] Furthermore, the development of capabilities that create credible fake videos, such as deepfakes, and information operations that spread false information, can easily take advantage of cognitive biases[20] and other structural constraints of human cognition, such as restricted working memory, attentional constraints, cognitive and confirmation bias, and could result in social harm, such as the diminishing importance of objective facts and data.[21] Moreover, privacy and human rights issues may arise in the context of AI-assisted targeting.
For example, reporting on the conflict in Gaza indicates that the Lavender AI targeting system used by the Israel Defence Forces identified roughly 37,000 individuals allegedly linked to terrorist groups. While human analysts did review the system’s recommendations before authorising strikes, they reportedly spent only around twenty seconds validating its outputs, often treating them as more reliable and leading, encouraging, and amplifying the human analysts’ will to let the machine effectively take over their “thinking”.[22] While the targeting process begins earlier through intelligence collection, the definition of selection criteria, and the tasking of operational assets, the final stage of human–machine interaction exposes important vulnerabilities. In practice, the operator’s role may be reduced to a time-limited veto rather than meaningful deliberation, simultaneously making human participation minimal and complex.[23] Such conditions can reinforce automation and action bias, encouraging operators to defer to algorithmic recommendations. This is particularly problematic where lethal force is involved, as targeted killings should require careful and deliberate decision-making. Even when formal decision-making frameworks exist, they may prove ineffective if operators rapidly confirm system outputs or if decision-makers disregard contradictory intelligence due to confirmation bias. Furthermore, machine-learning systems trained on biased or incomplete datasets risk discriminatory outcomes, while maintaining sufficiently updated and context-rich data in dynamic conflict environments remains a significant operational challenge.[24]
Legal Challenges Under International Humanitarian Law and the Law of Armed Conflict
The goal of this subsection is to determine whether international law contains guidelines that can educate and give a direction to nation-states as to how to operate autonomous systems in accordance with ethical principles. Although international discussions and experts have not adopted a generally accepted definition of autonomous weapon systems, with as many as 12 definitions,[25] a key question regarding the concept of weapon autonomy remains. Different states and international organisations interpret autonomy by assessing the levels of significance of legal, ethical, or military issues.[26] Despite this lack of consensus, it is still clear that the use of AI poses challenges for traditional concepts of accountability, attribution of responsibility and the requirement of meaningful human control over the use of force. The difficulty is that regulatory frameworks must take into account the speed of development of autonomous systems, which has outpaced existing traditional ethical and legal concepts and is outrunning the establishment of new ones, as well as the diverse global geopolitical and sociocultural backgrounds and philosophies.[27] As the dominating European, North American, and Asian ethical frameworks are neither uniform nor universally shared, and often reflect particular philosophical, political, or cultural traditions, regulatory guidelines derived primarily from them risk lacking global legitimacy.[28] Consequently, ethical norms embedded in international governance mechanisms for LAWS may conflict with or marginalise alternative religious, cultural, or societal value systems, complicating efforts to establish broadly accepted standards.[29] These aspects, based on ethical considerations, serve as the foundation for long-standing debates about the use of autonomous weapons in military contexts under IHL, particularly with respect to substantive principles, notably distinction, proportionality, and precaution.[30]
The Law of Armed Conflict (LOAC), also known as IHL, is the set of rules of international law, customary or conventional, intended to govern matters arising in situations of international or non-international armed conflict. LOAC is based on formal treaties such as the Hague and Geneva Conventions and Additional Protocols, as well as customary state practices and the Martens Clause, which extends protection beyond codified law based on “principles of humanity and the dictates of public conscience”.[31] Furthermore, article 36 of the Geneva Convention specifically states that in the “study, development, acquisition or adoption of a new weapon, means or method of warfare”, state parties to the convention are “under an obligation to determine whether its employment would, in some or all circumstances, be prohibited by this Protocol or by any other rule of international law”.[32] As AI integration into military systems could well be understood as a new weapon, states and international organisations have sparked a debate on the legal review of AI-reliant systems to ensure their use complies with LOAC.
In this context, three core principles of LOAC are particularly relevant. First, the principle of distinction requires a clear distinction between combatants and civilians. Second, the principle of proportionality requires an assessment of the use of force, which should be proportional to the military objective, therefore prohibiting attacks expected to cause excessive harm. Third, the principle of precaution requires taking steps to verify that the target is lawful and to ensure that it is not causing civilian harm.[33]
Critics, such as Human Rights Watch and International Human Rights Clinic in Losing Humanity: The Case Against Killer Robots, argue that fully autonomous weapons cannot comply with LOAC principles due to their inability to distinguish between combatants and non-combatants or assess proportionality in complex, dynamic environments.[34] For instance, it is unlikely that fully autonomous weapons are able to distinguish between “a fearful civilian and a threatening enemy” or even a surrendering enemy,[35] especially in the contexts of urban combat, which could see belligerents “ditch” their uniform and distinctive insignias for civilian clothing or seek shelter inside schools, hospitals and residential buildings, unjustly endangering civilians. Furthermore, fully autonomous weapons may not be able to meet the principle of proportionality, which demands a subjective analysis of the potential harm to civilians and collateral damage against the importance of the military objective on a case-by-case basis. Judging proportionality “requires more than a balancing of quantitative data” and involves an ethical, qualitative, and evaluative assessment by a human who weighs and compares complex values and maintains the moral essence and respect for human dignity.[36] The argument that AI systems may violate LOAC extends to AI-enabled cyber capabilities that may target civilian networks or systems.
Policy Implications and Governance Approaches
The ethical and legal challenges outlined in the previous section also have significant implications for policy and governance. As such, when considering the use of AI in warfare, policymakers must take into account ethical ramifications and compliance with existing legal frameworks. In this regard, governance approaches should address accountability gaps, clearly define responsibility, and ensure that AI deployment aligns with existing legal frameworks such as IHL.
The Need for a Coherent Governance Framework
Amongst other weapon systems, LAWSs fundamentally lack ethical legitimacy and, in many respects, legal legitimacy within the framework of IHL. Nevertheless, the current legal framework does not comprehensively address all ethical concerns at stake, creating a normative gap that has drawn increasing attention from governments, international organisations and policy institutes, which play a central role in shaping emerging governance frameworks.[37] Reports and policy statements from bodies such as the United Nations with its Group of Governmental Experts (UN GGE),[38] the NATO Secretary General, the ICRC, and national defence ministries have highlighted the urgency of establishing clearer rules and oversight mechanisms to regulate LAWS’ development and deployment before the end of 2026.[39] However, no coherent and comprehensive global governance framework currently exists; regulation is fragmented, as this section illustrates by examining three prominent approaches to mitigate ethical and legal challenges associated with AWS.
The difficulty in creating a regulatory framework that is both subject-matter and geographically comprehensive centres on the current avenues of deliberation and the positions of the major countries developing AI systems.[40] Until recently, the most prominent forum has been the UN’s GGE, which has seen the rise of 2 blocks on the current state of affairs on the regulation of LAWSs. Firstly, there is the coalition to ban LAWSs, which is constituted by NGOs and IOs, as well as most countries from the Global South and the People’s Republic of China.[41] However, the Chinese stance can be considered only “qualified support” for the ban, as their definition of what constitutes LAWs is very narrow and provides no monitoring mechanisms or sanctions for violations.[42] On the other hand, the United States, the Russian Federation, and several members of NATO generally maintain that such systems are already adequately regulated and caution that binding restrictions would be premature and potentially harmful to technological innovation.[43]
However, as of 2023, other than the establishment of 11 non-binding Guiding Principles on LAWSs, the progress towards the global regulation of AWS has stalled, in part due to the consensus-based format on which the GGE operates.[44] In response, alternative governance initiatives have emerged outside this framework, such as the 2023 First Summit on Responsible AI in the Military Domain (REAIM). This Summit produced a Blueprint for Action, which was organised and supported by a coalition of middle and smaller powers, including the Netherlands, the Republic of Korea, and the United Kingdom, calling for military AI to comply with international law and remain human-centric.[45] Notably, major powers such as the People’s Republic of China (PRC) and Russia have not endorsed the initiative, reflecting broader geopolitical competition over military AI.[46] Parallel normative pressure has also emerged from humanitarian actors such as the ICRC and the UN system, whose leadership has urged states to agree on international rules for autonomous weapons by 2026. Similarly, as of 2021, the EU Parliament adopted 2 resolutions on LAWS and military AI, and the UN General Assembly adopted its first resolution on autonomous weapons in 2023 and followed it with the 2024 resolution “Artificial Intelligence in the Military Domain and its Implications for International Peace and Security,” adopted by an overwhelming majority of member states.[47] Together, these initiatives suggest an emerging trend toward broader multilateral engagement, often driven by middle powers, seeking alternative pathways for governance at a time when major technological powers remain reluctant to accept binding constraints on military AI.
One of the most popular approaches to address the ethical challenges posed by fully autonomous AI systems is to maintain a human-in-the-loop.[48] This refers to the idea that a human “in the loop” can ensure compliance with laws and rules of engagement and be held accountable for the system’s actions. However, in some military applications, such as missile-interception systems, there is pressure to move humans into positions “on” the loop to avoid slowing the system’s reaction time. This raises questions about the extent to which a human can intervene rapidly enough to curtail an engagement.[49] The loop concept provides an approach to mitigating risk, but it is incomplete, as it doesn’t address the role of humans in other parts of the system life cycle (for instance, loitering weapons that, after launch, remain on stand-by until they identify a target, requiring a human to launch, satisfying the human in the loop requirement) or distinguish different varieties of command and control arrangements.
A further approach supports the idea of “meaningful human control”, a concept that aims to ensure human involvement in autonomous systems to mitigate risks. It requires humans to have significant moral, legal, and operational control over the system throughout its development and deployment, including design, training, and accountability structures. The U.S. Department of War prefers the concept of “appropriate human judgment”, which emphasises the importance of the human-machine relationship in the development and use of a system, rather than merely at the moment of decision-making, to further increase autonomy.[50] This is rooted in the belief that increased autonomy in weapon systems can better effectuate commanders’ intentions and that a ban on LAWS would undermine humanitarian values.[51] However, what constitutes “appropriate judgement” or “meaningful control” is still unclear, and it risks obscuring the challenges of human-machine interaction and being adversely shaped by ongoing state practices.[52] Although different commentators have proposed their own terminology for the type of human involvement required in autonomous weapons, there is a consensus that weapons must remain under human control.[53]
In previous UN CCW meetings, the GGE had identified various stages of the life cycle of weapon systems where human influence and control are necessary,[54] an idea that seems to have gained traction as the Working Paper for the 2026 GGE of the UN CCW emphasises the centrality of human control and responsibility over weapon systems.[55] By defining the phases of the system’s life cycle in which humans need to be involved, this approach emphasises the importance of human-led risk management throughout the life cycle, including post-deployment analysis. It portrays human involvement as a complex concept, with humans playing different roles at different times, not just during target selection and engagement, with the goal of ensuring risk management. To ensure human involvement and mitigate risk, some restrictions may be applied to the AI system’s design and development. A starting point could be the limitation on the time autonomous systems can operate without human involvement or direction. While some autonomous systems are designed to operate for extended periods, the “Autonomy in Weapon Systems” Directive from the U.S. Department of War states that autonomous weapons must complete their tasks within a timeframe aligned with the commander’s intentions.[56] If they are unable to do so, they must either terminate the task or seek additional human input before proceeding. In the event of malfunctions or when the system exceeds these constraints, it may be programmed to abort the mission and return to base.[57]
The absence of a unified framework, as highlighted by these three approaches presented, underscores the importance of proactive oversight and adaptive yet effective regulation, given the rapid pace of technological development and the high stakes of such military applications. As the UN Secretary General has emphasised in the past, the adoption of a legally binding instrument requiring meaningful human control is essential as “[w]e cannot delegate life-or-death decisions to machines [and] time is running out to take preventative action”.[58] Accordingly, there is a need for a coherent regulatory framework that can evolve alongside technological advances while also providing clear, enforceable standards to ensure accountability and adherence to ethical values and legal norms.
Towards Responsible Autonomy in Warfare
To address the broadening capabilities of AI systems, there may be limitations on the number and types of tasks they can perform without human involvement. However, as remarked by the growing number of standards, assessment templates, review processes, and other tools and good practices that aim to operationalise the responsible adoption of AI;[59] autonomous systems that pose more significant risks, such as those capable of lethal actions or targeting people, may require additional controls and testing for safety purposes.[60] To ensure that the system functions as intended, reliability and predictability standards should be established. These standards must be established and met during the system’s design and development phase and must be continuously maintained, with periodic evaluations. It is reasonable to expect that more impactful systems will require greater reliability and predictability. A similar aspect of system design to be considered to help in mitigating the risk of AI is the ability to intervene and stop the system, especially in case of accidents and malfunctions, where the operator should be able to intervene in a timely manner to redirect the system’s actions if necessary, while still meeting mission requirements. For systems using lethal force in operations near civilians, operators should be able to intervene quickly; when operating systems that do not use lethal force or those taking defensive actions in areas far from civilians, the intervention time may be slower but still reasonable.[61]
Another aspect to consider is the level of transparency required for the system. To reduce risks associated with AI, the operator must know what the system is designed to do in a specific operational context and understand how the system made critical decisions or took specific actions.[62] Moreover, a best practice for operating autonomous systems is to use an interface that is easy for trained operators to understand, provides feedback on system status, and includes clear procedures for activating and deactivating system functions.[63] The chair’s summary of the April 2018 UN CCW GGE meeting also emphasised the importance of operators knowing the characteristics of the weapon system and having sufficient and reliable information to make informed decisions and ensure legal compliance.[64]
While current discourse often assumes that the future of military decision-making will retain a “human in the loop” and that the optimal approach is to pair AI with human operators in human–machine teams, such assumptions may prove only partially valid, as human oversight in such systems can become nominal in practice, and that the balance of decision authority between humans and autonomous systems may shift depending on system design or operational tempo.[65] In the short term, these models are likely to dominate; however, psychological research highlights the inherent limitations of human cognition, including restricted working memory, attentional constraints, cognitive and confirmation bias, and loss aversion, which, in certain contexts, can render automated analysis superior to human reasoning, particularly when information flows at increasing speed.[66] This raises critical operational questions, such as whether automating processes like Positive Identification (PID) or target analysis could empirically reduce error rates compared to human-in-the-loop systems. Consequently, defence strategies must not only focus on promoting effective human–machine partnerships but also identify domains in which human cognitive limitations may become the principal constraint on AI performance. This will define where human judgment must remain paramount and where AI deployment may be most appropriate, especially in high-tempo combat environments where adversaries may fully exploit the capabilities of their autonomous systems.[67]
At present, the technologies deployed remain relatively rudimentary compared to the ambitions of leading militaries. Nonetheless, advancements in AI offer significant opportunities for military operations, potentially reducing personnel risk and increasing precision compared to conventional forces. These benefits, however, are inseparable from profound ethical challenges, and it is therefore necessary to ensure the responsible use of AI in the military domain. This requires balancing operational effectiveness with ethical principles, accountability, and human oversight. To this end, governments should prioritise the organisation, training, and equipping of armed forces to operate effectively in a security environment where AI-enabled systems play a decisive role across multiple domains. Equally important is the development of comprehensive frameworks to address the ethical issues raised by human rights advocates and international bodies such as the United Nations. Only through such measures can the integration of AI into military systems proceed in a manner that maximises its strategic advantages while safeguarding the moral and legal standards that underpin legitimate military action.
Conclusion
Artificial intelligence is rapidly reshaping military operations and decision-making, creating ethical, legal, and governance challenges that existing regulatory frameworks only partially address. While AI-enabled systems may enhance operational effectiveness, precision, and force protection, they also risk weakening accountability, amplifying automation bias, and reducing meaningful human judgment in the use of force. The principal challenge, therefore, lies not simply in the autonomy of the technology itself but in the evolving relationship between human decision-makers, AI systems, and the institutional frameworks governing their use.
This paper has argued that maintaining formal human involvement alone is insufficient if operational conditions reduce oversight to little more than nominal approval. As military organisations increasingly integrate AI into targeting, intelligence analysis, and autonomous systems, ensuring compliance with international humanitarian law will depend on preserving meaningful human deliberation, transparency, and accountability throughout the entire system life cycle.
Ultimately, the integration of AI into warfare is likely inevitable. The critical question for policymakers is therefore not whether AI should be used in military operations, but under what legal, ethical, and governance conditions its use can remain compatible with human responsibility, international humanitarian law, and the principles that underpin legitimate military action.
[1] Ivana Zirojević, “Use of Artificial Intelligence in Contemporary Wars,” Vojno Delo 76, no. 1 (2024): 83–85, https://doi.org/10.5937/VOJDELO2401073Z 
.
[2] Terilyn Johnston Huntington and Amy E. Eckert, “‘We Watched His Whole Life Unfold… Then You Watch the Death’: Drone Tactics, Operator Trauma, and Hidden Human Costs of Contemporary Wartime,” International Relations 36, no. 4 (2022): 640, https://doi.org/10.1177/00471178221135036 ; John R. Emery and Hadley Briggs, “Human, All Too Human: Drones, Ethics, and the Psychology of Military Technologies,” Political Psychology 43, no. 3 (2022): 609–10, https://doi.org/10.1111/pops.12809 .
[3] Gábor Farkas and Gábor Fazekas, “Potential Health Risks for FPV Drone Operators,” Honvédorvos 76, KSZ (2024): 46–53, https://doi.org/10.29068/HO.2024.ksz.46-55 ; Huntington and Eckert, “‘We Watched His Whole Life Unfold,’” 639.
[4] Emery and Briggs, “Human, All Too Human,” 610; Huntington and Eckert, “‘We Watched His Whole Life Unfold,’” 648.
[5] Constant and long-term monitoring of potential targets, evaluation of collateral risks, and assessment of attack effects create a sense of intimacy between operator and target. While drones dramatically expand the physical distance separating combatants, they also enable operators to observe targets’ daily routines over prolonged periods, intensifying psychological engagement and mitigating the dehumanising effects traditionally associated with long-range killing. In this sense, spatial distance is traded for temporal proximity, creating a unique emotional and ethical burden. See Emery and Briggs, “Human, All Too Human,” 609; Huntington and Eckert, “‘We Watched His Whole Life Unfold,’” 640.
[6] Emery and Briggs, “Human, All Too Human,” 607.
[7] James Johnson, “Can AI Behave Ethically during Military Crises?,” International Affairs 102, no. 1 (2026): 68ff., https://doi.org/10.1093/ia/iiaf191 .
[8] Marta Bistron and Zbigniew Piotrowski, “Artificial Intelligence Applications in Military Systems and Their Influence on the Sense of Security of Citizens,” Electronics 10, no. 871 (2021): 14–15, https://doi.org/10.3390/electronics10070871 .
[9] Alexey Turchin and David Denkenberger, “Military AI as a Convergent Goal of Self-Improving AI,” in Artificial Intelligence Safety and Security, ed. Roman V. Yampolskiy (New York: Chapman and Hall/CRC, 2018), 2, https://doi.org/10.1201/9781351251389 .
[10] Satyadhar Joshi, “Agentic AI for National Defense: Review of Architectural Patterns and Recent Developments,” TechRxiv (October 2025): 1, 5–6, https://doi.org/10.36227/techrxiv.176045912.25605018/v1 ; Satyadhar Joshi, “Agentic Generative AI and National Security: Policy Recommendations for US Military Competitiveness” (2025): 1, http://dx.doi.org/10.2139/ssrn.5529680 .
[11] Joshi, “Agentic AI for National Defense,” 7–8.
[12] Daniel Trusilo, “Autonomous AI Systems in Conflict: Emergent Behaviour and Its Impact on Predictability and Reliability,” Journal of Military Ethics 22, no. 1 (2023): 3–5, https://doi.org/10.1080/15027570.2023.2213985 ; Aleksandra Seizovic, David Thorpe, and Steven Goh, “Emergent Behavior in the Battle Management System,” Applied Artificial Intelligence 36, no. 1 (2022): 4–6, https://doi.org/10.1080/08839514.2022.2151183 .
[13] Doug Bonderud, “AI Decision-Making: Where Do Businesses Draw the Line?,” IBM, January 31, 2025, https://www.ibm.com/think/insights/ai-decision-making-where-do-businesses-draw-the-line.
[14] Mary L. Cummings, “Creating Moral Buffers in Weapon Control Interface Design,” IEEE Technology and Society Magazine 23, no. 3 (2004): 29–30; Lucy Dibsdale, “Beyond Human Judgment – The Morality of Machines: A Critical Examination of the Ethical and Moral Dimensions of Autonomous and AI Weapon Systems in Modern Warfare,” Journal of Global Faultlines 12, no. 2 (2025): 131, 133.
[15] Johnson, “Can AI Behave Ethically during Military Crises?,” 68ff.; Zirojević, “Use of Artificial Intelligence in Contemporary Wars,” 83–84.
[16] United Nations, “Universal Declaration of Human Rights” (1948).
[17] Christof Heyns, “Autonomous Weapon Systems and Human Rights Law,” presentation at the informal expert meeting organised by the state parties to the Convention on Certain Conventional Weapons, Geneva, May 13–16, 2014; U.S. Congress, Committee on Armed Services, “Hearing to Consider the Nomination of General Paul J. Selva, USAF, for Reappointment to the Grade of General and Reappointment to Be Vice Chairman of the Joint Chiefs of Staff,” Washington, DC, 2017.
[18] Jie Guo, “The Ethical Legitimacy of Autonomous Weapons: Reconfiguring War Accountability in the Age of Artificial Intelligence,” Ethics & Global Politics 18, no. 3 (2025): 30ff.
[19] Bistron and Piotrowski, “Artificial Intelligence Applications in Military Systems,” 12–15.
[20] The way a particular person understands events, facts, and other people, based on their own beliefs and experiences, which may not always be reasonable or accurate.
[21] Keith Dear, “Artificial Intelligence and Decision-Making,” RUSI Journal 164, nos. 5–6 (2019): 22, https://doi.org/10.1080/03071847.2019.1693801 .
[22] Peter Layton, “Artificial Intelligence at War,” Australian Strategic Policy Institute, August 20, 2024, https://www.aspistrategist.org.au/artificial-intelligence-at-war/. continue, please. do them all [23] Ingvild Bode, Hendrik Huelss, Anna Nadibaidze, Guangyu Qiao-Franco, and Tom F. A. Watts, “Prospects for the Global Governance of Autonomous Weapons: Comparing Chinese, Russian, and US Practices,” Ethics and Information Technology 25, no. 5 (2023): 3–4, https://doi.org/10.1007/s10676-023-09678-x .
[24] Nicolae Sfetcu, “The Evolution of AI,” in Artificial Intelligence in Intelligence Agencies (Bucharest: MultiMedia Publishing, 2024), 17ff.; Najm A. Kh. Alhatimi Aleessawi, “AI-Powered Warfare: Navigating the Strategic, Ethical, and Geopolitical Frontiers of Autonomous Arms Races,” Journal of Strategic Studies and Political Research 4, no. 1 (2025): 165; Bode et al., “Prospects for the Global Governance of Autonomous Weapons,” 3; Johnson, “Can AI Behave Ethically during Military Crises?,” 76; Zirojević, “Use of Artificial Intelligence in Contemporary Wars,” 84.
[25] See, for instance, United States Department of Defense and European Parliament resolutions as cited in Katerina Yordanova, “Artificial Intelligence and Armed Conflicts,” in The Law, Ethics and Policy of Artificial Intelligence, ed. Nathalie A. Smuha (Cambridge: Cambridge University Press, 2025), 423.
[26] Zirojević, “Use of Artificial Intelligence in Contemporary Wars,” 81–84.
[27] Chris Gilbert and Mercy Abiola Gilbert, “The Security Implications of Artificial Intelligence (AI)-Powered Autonomous Weapons: Policy Recommendations for International Regulation,” International Research Journal of Advanced Engineering and Science 9, no. 4 (2024): 210; Najm A. Kh. Alhatimi Aleessawi, “AI-Powered Warfare,” 162.
[28] These are the global regions leading AI development, leaving South America and Africa comparatively underrepresented.
[29] Cathy Roche, P. J. Wall, and Dave Lewis, “Ethics and Diversity in Artificial Intelligence Policies, Strategies and Initiatives,” AI and Ethics 3 (2023): 1096–97, 1100.
[30] Rebecca Crootof, “War, Responsibility, and Killer Robots,” North Carolina Journal of International Law 40 (2015): 928ff.; Michael Schmitt, “Unmanned Combat Aircraft Systems and International Humanitarian Law: Simplifying the Oft Benighted Debate,” Boston University International Law Journal 30 (2012): 609–17.
[31] Rupert Ticehurst, “The Martens Clause and the Laws of Armed Conflict,” International Review of the Red Cross 37, no. 317 (April 1997): 128.
[32] International Committee of the Red Cross, “Article 36 – New Weapons,” in Protocols Additional to the Geneva Conventions of 12 August 1949 and Relating to the Protection of Victims of International Armed Conflicts (Protocol I), June 8, 1977.
[33] Forrest E. Morgan et al., Military Applications of Artificial Intelligence: Ethical Concerns in an Uncertain World (Santa Monica, CA: RAND Corporation, 2020), 31ff.
[34] Bonnie Docherty, Losing Humanity: The Case Against Killer Robots (Human Rights Watch, 2012).
[35] Sebastian Clapp, Defence and Artificial Intelligence (European Parliamentary Research Service, April 2025), 9, https://www.europarl.europa.eu/RegData/etudes/BRIE/2025/769580/EPRS_BRI(2025)769580_EN.pdf; Morgan et al., Military Applications of Artificial Intelligence, 31.
[36] Morgan et al., Military Applications of Artificial Intelligence, 31.
[37] Eugenio V. Garcia, “Artificial Intelligence, Peace and Security: Challenges for International Humanitarian Law,” SSRN (2019): 5, https://ssrn.com/abstract=3595340.
[38] The UN CCW GGE stands for the United Nations Convention on Certain Conventional Weapons – Group of Governmental Experts.
[39] Cengiz Özbek and Tolga Erdem, “Technological Trends in Security Policies of NATO: Artificial Intelligence (AI),” Journal of Mehmet Akif Ersoy University Economics and Administrative Faculty 12, no. 4 (2025): 1555; Denise Garcia, “The Global Diplomacy of Governing Military Artificial Intelligence,” Ethics and Information Technology 27, no. 55 (2025): 5, https://doi.org/10.1007/s10676-025-09863-0 .
[40] Gilbert and Gilbert, “The Security Implications of Artificial Intelligence (AI)-Powered Autonomous Weapons,” 210.
[41] Nik Hynek and Anzhelika Solovyeva, “Operations of Power in Autonomous Weapon Systems: Ethical Conditions and Sociopolitical Prospects,” AI and Society 36 (2021): 80, https://doi.org/10.1007/s00146-020-01048-1 .
[42] Bode et al., “Prospects for the Global Governance of Autonomous Weapons,” 3; Viacheslav Biletskyi and Serhii Khalymon, “General Artificial Intelligence and the US–PRC Arms Race: Impacts on Global Militarization and International Law,” Revista Juridica Portucalense 1, no. 39 (2026): 68, https://doi.org/10.34625/issn.2183-2705(39.1)2026.ic-3 .
[43] Michael L. Littman et al., Gathering Strength, Gathering Storms: The One Hundred Year Study on Artificial Intelligence (AI100) 2021 Study Panel Report (Stanford, CA: Stanford University, September 2021), 41; Bode et al., “Prospects for the Global Governance of Autonomous Weapons,” 3.
[44] Hynek and Solovyeva, “Operations of Power in Autonomous Weapon Systems,” 93; Bode et al., “Prospects for the Global Governance of Autonomous Weapons,” 2.
[45] Garcia, “The Global Diplomacy of Governing Military Artificial Intelligence,” 5.
[46] Biletskyi and Khalymon, “General Artificial Intelligence and the US–PRC Arms Race,” 68.
[47] Garcia, “The Global Diplomacy of Governing Military Artificial Intelligence,” 5; Clapp, Defence and Artificial Intelligence, 11.
[48] Observe–orient–decide–act; Morgan et al., Military Applications of Artificial Intelligence, 12, 16, 41.
[49] Garcia, “Artificial Intelligence, Peace and Security,” 4, 11.
[50] U.S. Department of Defense, Autonomy in Weapon Systems, DoD Directive 3000.09 (Washington, DC: Department of Defense, 2023), 3.
[51] U.S. Delegation Statement on Possible Options, Meeting of the Group of Governmental Experts of the High Contracting Parties to the CCW on Lethal Autonomous Weapons Systems, Geneva, 2018; Clapp, Defence and Artificial Intelligence, 4.
[52] Lena Trabucco, “What Is Meaningful Human Control, Anyway? Cracking the Code on Autonomous Weapons and Human Judgment,” Modern War Institute at West Point, September 21, 2023, https://mwi.westpoint.edu/what-is-meaningful-human-control-anyway-cracking-the-code-on-autonomous-weapons-and-human-judgment/; Bode et al., “Prospects for the Global Governance of Autonomous Weapons,” 4.
[53] Clapp, Defence and Artificial Intelligence, 9–11.
[54] United Kingdom, “Human Machine Touchpoints: The United Kingdom’s Perspective on Human Control over Weapon Development and Targeting Cycles,” submission to the CCW/GGE.2 on Other Matters, August 8, 2018.
[55] CCW/GGE.1/2026/WP.1, Working Paper Submitted by Egypt, 2–3; Yordanova, “Artificial Intelligence and Armed Conflicts,” 422ff.
[56] U.S. Department of Defense, Autonomy in Weapon Systems, 4.
[57] For instance, limiting the operational range of an AI system may be another useful restriction, as certain autonomous systems possess advanced navigation capabilities enabling them to move freely through the air. To minimise associated risks, operators could establish geographic boundaries restricting autonomous movement.
[58] United Nations, “Lethal Autonomous Weapon System ‘Politically Unacceptable, Morally Repugnant and Should Be Banned’, Secretary-General Says during Informal Consultations on Issue,” UN Meetings Coverage and Press Releases, May 12, 2025, https://press.un.org/en/2025/sgsm22643.doc.htm; United Nations, “‘Politically Unacceptable, Morally Repugnant’: UN Chief Calls for Global Ban on ‘Killer Robots,’” UN News, May 14, 2025, https://news.un.org/en/story/2025/05/1163256.
[59] NATO, “Summary of NATO’s Revised Artificial Intelligence (AI) Strategy,” July 10, 2024, https://www.nato.int/en/about-us/official-texts-and-resources/official-texts/2024/07/10/summary-of-natos-revised-artificial-intelligence-ai-strategy.
[60] Özbek and Erdem, “Technological Trends in Security Policies of NATO,” 1551.
[61] Zirojević, “Use of Artificial Intelligence in Contemporary Wars,” 89.
[62] Bistron and Piotrowski refer to this as the “right to explanation,” meaning that an operator should be able to receive an explanation for an algorithmic output when it is not immediately intuitive or understandable.
[63] United States, “Autonomy in Weapon Systems,” submission to the CCW/GGE.1 on the Examination of Various Dimensions of Emerging Technologies in the Area of Lethal Autonomous Weapons Systems, November 10, 2017; U.S. Department of Defense, Autonomy in Weapon Systems, 11–16.
[64] Chair’s Summary of the Discussion on Agenda Items 6(a)(b)(c)(d), Group of Governmental Experts on Emerging Technologies in the Area of Lethal Autonomous Weapons Systems, 2018.
[65] Mikey Dickerson, “The Military’s Fabled ‘Human in the Loop’ for AI Is Dangerously Misleading,” DefenseNews, March 26, 2026, https://www.defensenews.com/opinion/2026/03/26/the-militarys-fabled-human-in-the-loop-for-ai-is-dangerously-misleading/; Peter Layton, “Human-Machine Teaming’s Shared Cognition Changes How War Is Made,” RUSI, January 27, 2025, https://www.rusi.org/explore-our-research/publications/commentary/human-machine-teamings-shared-cognition-changes-how-war-made/.
[66] Dear, “Artificial Intelligence and Decision-Making,” 22–23.
[67] Ajai Raj, “Generative AI Wargaming Promises to Accelerate Mission Analysis,” Johns Hopkins Applied Physics Laboratory, March 3, 2025, https://www.jhuapl.edu/news/news-releases/250303-generative-wargaming.
