Executive Summary
- AI functions as an enabling technology in military contexts, accelerating decision-making, automating targeting, and reshaping command and control structures.
- AI systems risk undermining the human judgment required to comply with IHL’s core principles of distinction, proportionality, and precaution. Moreover, there exists a strong risk of inherent biases in such systems, resulting in the disproportionate or indiscriminate engagement of targets.
- Lethal Autonomous Weapons Systems (LAWS) are inherently unpredictable in volatile conflict environments, making IHL compliance structurally difficult to guarantee.
- Existing legal frameworks are insufficient in governing the use of AI and LAWS in warfare. As such, the following recommendations are made: (1) establish binding international legal frameworks for AI and LAWS, (2) mandate IHL compliance reviews, (3) require algorithmic bias audits, and (4) ensure human control and accountability.
Introduction
The recent Pentagon decision to expand the use of artificial intelligence (AI) in US military operations has brought increased attention to the use of emerging technologies in warfare and has raised a number of ethical and operational concerns (Dorsey et al., 2026). In addition, the deployment of AI in modern warfare has raised questions regarding its compatibility with international humanitarian law (IHL). Particularly, the growing development of AI-supported Lethal Autonomous Weapons Systems (LAWS) highlights the need for efforts to delineate the legal framework and limitations of lethal military decision-making in autonomous systems.
AI in Modern Warfare
The use of AI in the military domain must be viewed not as a distinct weapon but as an enabling technology that impacts the way in which modern warfare is conducted (Scharre, 2020). Today, these technologies are increasingly embedded in military operations, thus serving as a catalyst for increased automation, accelerated decision-making, and improved command and control (Sauer, 2022). The immediate processing of image and video data can offer opportunities for militaries to improve targeting through the automated recognition of potential targets (Tucker, 2019). Such technologies can be deployed in all areas of the military realm, both on the offensive and the defensive (Fischer, 2022). Recent reports by human rights organisations have exposed how the Israel Defense Forces (IDF) have used AI to help identify and accelerate decision-making in its military operations in Gaza (Human Rights Watch, 2024). Such practices have been criticised extensively due to their inaccuracy, lack of consideration for human dignity, processing speed, and their association with violations of IHL. The most controversial application of AI lies in Lethal Autonomous Weapons Systems (LAWS), which can autonomously identify and lethally target individuals or other potential military objectives (Fischer, 2022).
The military applications of AI have raised various concerns. First, the use of automated and unmanned weapon systems raises questions regarding their level of autonomy (ICRC, 2019). There is a risk of losing human control as the essential source of judgment needed to comply with IHL. In this regard, the International Committee of the Red Cross (ICRC) has been concerned with whether an algorithm is able to substitute human intelligence in distinguishing between combatants and non-combatants – a highly context-specific distinction (ICRC, 2018). For example, would AI be able to recognise a combatant that surrenders themselves, thus changing their status under IHL?
Similarly, the ICRC has raised concerns regarding the reduced latency associated with AI pattern recognition (ICRC, 2018, 2019). Where an operator previously analysed intelligence on potential targets at a human cognition speed, they can now receive exponentially larger amounts of AI-designated targets for which they must decide whether or not to engage. In the case of the IDF’s Lavender programme (which holds an acceptable misidentification rate of 10 %), it has become clear that AI has not been used to improve the accuracy of targeting, but rather both to expand its scope of capabilities and to increase the frequency at which supposed combatants can be targeted (Human Rights Watch, 2024; Sylvia, 2024).
Compatibility With International Humanitarian Law
As such, these technologies raise questions regarding their compatibility with international humanitarian law (IHL) (ICRC, 2019). In general, it is understood that, as an enabling means of warfare, such technology must be used in compliance with the principles of IHL. Therefore, military operators must (1) distinguish between lawful military targets and civilians or civilian infrastructure, (2) ensure that such use of force is proportional to the military objective to be achieved, and (3) take precautionary measures to minimise collateral damage. Hence, in cases where AI algorithms are used in combat (physical or in cyberspace), their design must ensure the ability of operators to assess an attack’s compliance with these core principles.
However, when AI systems serve to enable human decision-making, the accelerated data processing speed can blur human judgment (Saxon, 2024). Advances in AI decrease the latency period between when algorithms receive and process data and the moment this intelligence can be assessed by a human operator. As such, significantly more potential targets can be identified in a shorter time span, which can increase pressure on individual military actors to make calculated decisions that comply with IHL. Hence, the operator is faced with beyond-human cognition speeds, leaving less time for accurate assessment of the legal status of a designated target (Schuller, 2017).
While still in its infancy, recent technological innovations have accelerated the development of Lethal Autonomous Weapons Systems (LAWS), which often run on AI (Shah, 2026). As such, important questions arise regarding their compatibility with IHL, particularly in relation to the unlawfulness of indiscriminate attacks. As AI in (L)AWS is inherently unpredictable, rendering it difficult for humans to anticipate the decision-making process by which they function, fundamental challenges for IHL emerge (ICRC, 2021). While the behaviour of AWS may be more controllable in strictly closed and monitored environments, this control is difficult to maintain in a volatile conflict environment – particularly considering the ever-evolving urbanisation and multidimensionality of modern conflicts (Bhuta and Pantazopoulos, 2016).
In addition to the lack of predictability, AI systems often suffer from systemic biases based on the biases of those who programmed and trained such technologies (Bode, 2024; Human Rights Watch, 2024; Ojha, 2025). As seen in Israel’s use of AI in Gaza, the bias in identifying supposed combatants and the accepted 10 % misidentification rate challenge the compatibility of such systems with IHL (Human Rights Watch, 2024). Such bias could allow for situations in which an AWS wrongfully engages supposed targets, in a disproportionate manner, and with a lack of consideration for potential collateral damage (Ojha, 2025).
While the core principles of IHL remain legal obligations, even when decision-making functions are delegated to an algorithm, regulatory gaps remain present (Davison, 2018). Currently, there is no sufficient legal framework to deal with the use of AI in warfare (Ojha, 2025). Under international law, accountability cannot be transferred to machines, weapons, or algorithms (Davison, 2018). Consequently, those involved in the design and deployment of AWS must ensure that these systems can be used in a manner consistent with IHL. This requirement is reflected in Article 36 of AP I, which obliges States Parties to review new weapons, means, and methods of warfare for their compatibility with IHL. (Protocol Additional to the Geneva Conventions of 12 August 1949, and relating to the Protection of Victims of International Armed Conflicts (Protocol I), of 8 June 1977, 1977).
Nonetheless, questions regarding the governance of such technologies remain unaddressed by international law (Davison, 2018). While there exists a general agreement that weapons systems must employ a level of “context-appropriate human control and judgement” (Group of Governmental Experts on Lethal Autonomous Weapons Systems, 2024, p. 1), the exact limitations of such systems’ autonomy remain opaque. However, when humans exert limited or no control over such systems, it will become increasingly difficult to ensure individual accountability under international criminal law, given that those who designed, trained, or activated a (L)AWS might not be aware of the full intent of such systems when they make independent targeting decisions (Davison, 2018).
Policy Recommendations
- Establish binding international legal frameworks for AI and LAWS
While the use of AI and LAWS in warfare are understood to fall completely under the principles of IHL, a lack of clear legal delineation and adaptation to modern technology has produced a significant regulatory gap. Therefore, states must proactively negotiate binding legal frameworks that regulate the use of AI in military applications and impose a global ban on LAWS (see UN News, 2025). While some efforts are being made to regulate (L)AWS, the difficulty in defining ‘human control’ over such systems highlights how an outright ban remains the safest way forward.
- Mandate IHL compliance reviews
As exists in other international legal frameworks regulating warfare, such as the Chemical Weapons Convention (CWC) and the Nuclear Non-Proliferation Treaty (NPT), legal frameworks on AI and (L)AWS should incorporate an independent compliance mechanism that monitors and reviews the compliance of new technologies with the core principles of IHL before their deployment.
- Require algorithmic bias audits
The incorporation of AI systems in military technologies raises concerns regarding inherent biases from those who developed and trained these systems. Any legal framework on the use of AI in warfare should thus include an independent auditing mechanism that particularly reviews demographic biases.
- Ensure human control and accountability
Considering legal accountability cannot be transferred to a machine, any legal framework on the use of AI in warfare must ensure the need for human control before deployment. As such, accountability will remain with the operating officer. In this regard, it is imperative that the reduced latency of AI data processing systems in military applications does not exceed human cognition speeds.