The Jurisprudence of Functional Identity: Artificial Intelligence as a Combatant in Warfare Under Article 43(2) of Additional Protocol I to the Geneva Conventions

Abstract

Article 43(2) of Additional Protocol I to the Geneva Conventions defines combatants but does not specify that they must be human. Bundle Theory, which views identity as a collection of rights and duties, suggests that legal personhood is unnecessary for combatant status. Artificial intelligence, while not a legal person, could meet the functional criteria of a combatant, such as Roe v Wade (1973), where personhood was contextually recognised. Similarly, artificial intelligence could be classified as a combatant under international as participating in hostilities and adhering to a command structure. This parallels Roe v. humanitarian law without being granted full legal personhood.

1. Introduction

Article 43(2) of Additional Protocol I to the Geneva Conventions (AP I) mentions that, ‘[m]embers of the armed forces of a Party to a conflict (other than medical personnel and chaplains covered by Article 33 of the Third Convention) are combatants, that is to say, they have the right to participate directly in hostilities’.

Read together with the adjoining provisions, Article 43(1) defines the armed forces of a Party as “all organised armed forces, groups and units which are under a command responsible to that Party for the conduct of its subordinates,” while Article 43(3) obliges Parties to “communicate to one another the names and designations of their armed forces.” This textual context shows that the drafters conceived of an organised, command-responsible structure rather than specifying the biological or artificial nature of its members.

The central objective of this article is to determine whether certain military AI systems may meet the functional criteria of combatant status under Article 43(2) of AP I without conferral of legal personhood. The aim is not to absolve humans of liability, but rather to preserve accountability by avoiding a “responsibility gap” should AI assume direct combat roles. Accordingly, this paper asks whether a narrow, functional combatant category—analogous to limited corporate personhood—may be doctrinally necessary to align IHL with emerging technologies without displacing human agency or command responsibility.

This raises a core research question:

Can IHL recognise AI as a functional combatant category solely for regulatory purposes—without granting it full legal personhood—so that human responsibility is preserved rather than displaced?

This question frames the doctrinal analysis that follows and helps clarify that this paper does not advocate for AI autonomy or moral agency, but seeks to determine whether existing IHL categories are sufficient, or whether new functional categories are needed to uphold responsibility. Accordingly, any proposed functional category for AI under Article 43(2) operates strictly to maintain existing command responsibility structures, not to transfer liability to AI systems.

Research Methodology

This study adopts a doctrinal and conceptual legal methodology grounded in treaty interpretation under Articles 31 and 32 of the Vienna Convention on the Law of Treaties (VCLT). Accordingly, it analyses the conditions under which AI systems engaged in hostilities could satisfy the functional criteria of combatant status under Article 43(2) of AP I.

This research proceeds in three stages.

First, it examines the textual and contextual interpretation of the relevant treaty provisions, Articles 43(1)–(3) of AP I and Article 4 of the Third Geneva Convention (GC III), by reference to ICRC Commentaries, national military manuals, and academic commentary.

Second, it undertakes a comparative doctrinal analysis of how different legal systems and scholarly interpretations have understood the notion of “membership in the armed forces,” identifying whether this concept is inherently anthropocentric or institutionally constructed.

Third, it applies the Bundle Theory of legal personhood to test whether combatant status can be attributed to non-human entities for functional and regulatory purposes, without conferring full legal personhood.

This research is qualitative and interpretive, not empirical. It relies on primary legal instruments (treaties, commentaries, manuals, and case law) and secondary scholarly literature to elucidate how legal concepts evolve through analogy and functional reasoning.

The methodology thus reflects a conceptual–normative approach aimed at clarifying how IHL may adapt to emerging technologies while preserving its humanitarian and accountability principles. The objectives articulated in this section are revisited in the Conclusion, which evaluates to what extent these objectives have been achieved and identifies areas for doctrinal development.

Having outlined the research objectives and methodological framework, the next section surveys existing scholarship on whether artificial intelligence could qualify as a combatant under international humanitarian law.

2. Perspectives of Artificial Intelligence as Combatants

The ambiguity surrounding whether Article 43(2) of AP I implicitly limits combatant status to human persons has prompted debate among scholars. Some have considered autonomous and AI technologies that are independent in decision-making as potential combatants (Roff 2015), while others have proposed that AI blurs the distinction between the concepts of means of warfare and combatants (Heyns 2016). Others, however, have refrained from categorising dehumanised technologies as either combatants or weapons under IHL, arguing that a new regulatory category is necessary to address such technologies, highlighting the insufficiency of existing legal frameworks.

In addition to Articles 43 of AP I and 4 of the Third Geneva Convention, several contemporary instruments and policy frameworks further illuminate the regulatory landscape governing the military use of artificial intelligence (see Alsamara and Farouk 2025). The United Nations Institute for Disarmament Research (UNIDIR) underscores that the military use of AI is broad, extending well beyond target engagement and weapons systems (Grand-Clément 2023) and that “militaries are increasing their use of AI across a range of applications and contexts,” (Government of The Netherlands 2023), including logistics, intelligence, surveillance, and reconnaissance. Similarly, the REAIM Call to Action, an intergovernmental initiative issued by the Government of The Netherlands (2023), recognises that “militaries are increasing their use of AI across a range of applications and contexts,” reflecting growing international concern over the governance of such technologies. At the national level, the U.S. Department of Defense Directive establishes policy on the design, development, and use of autonomous and semi-autonomous weapon systems, mandating that these operate under “appropriate levels of human judgment” (United States Department of Defense 2023). Together, these sources complement the treaty-based framework by demonstrating how states and international bodies are adapting interpretive and regulatory guidance to the emerging realities of military AI.

Comparable strategic documents adopted by other jurisdictions further illustrate the global breadth of regulatory engagement with military AI. The European Union’s Coordinated Plan on Artificial Intelligence Review (European Commission 2021) emphasises ethical oversight and dual-use governance; China’s New Generation Artificial Intelligence Development Plan (State Council of the People’s Republic of China 2017) explicitly identifies defence modernisation as a priority; and South Africa’s National Artificial Intelligence Policy Framework (Department of Communications and Digital Technologies (South Africa) 2024) highlights the importance of aligning AI innovation with ethical governance, human-centred design, safety and security, and socio-economic development. Collectively, these strategies confirm that the regulation of AI in the military sphere is emerging as a shared international concern rather than a solely Western policy initiative.

Beyond Western frameworks, China’s Position Paper on Regulating Military Applications of Artificial Intelligence (Government of the People’s Republic of China 2021) submitted to the UN Group of Governmental Experts on LAWS underscores the principle of “human control at all stages” of AI deployment and calls for a legally binding instrument to govern autonomous weapons. Taken together with the U.S. Department of Defense Law of War Manual and the European Union’s coordinated AI governance initiatives, these positions demonstrate a growing convergence on maintaining human accountability within AI-enabled military operations. This comparative alignment enhances the normative credibility of the functional combatant framework advanced in this paper.

Before assessing whether military AI technologies can be classified as combatants, it is essential to engage more deeply with the nature of AI itself. As recent scholarship on embodied AI and AI-enabled devices demonstrates, the term “artificial intelligence” encompasses both physically instantiated systems capable of environmental interaction and disembodied algorithmic architectures that support human decision-making (J. J. Bryson 2019; Beer 2015; Gunkel 2021). Embodied AI, such as robotic soldiers or autonomous drones, operates through sensors and effectors that permit autonomous engagement with the physical world, whereas AI-enabled systems remain tethered to human operators as decision-support tools. It is these specific operational features, rather than anthropomorphism, that determine whether an AI system could satisfy Article 43(2)’s continuous combat function requirement. This ontological distinction has direct legal significance: only embodied or functionally autonomous systems could, in principle, satisfy Article 43(2)’s requirement of continuous combat function. These features justify the exploration of an intermediate legal status: neither human nor “mere weapon”. AI’s embeddedness in command structures, its predictable operational function, and its capacity for battlefield execution create a functional role that existing categories of ‘means’ or ‘methods’ do not entirely capture under IHL. The purpose is not to liberate AI from human responsibility—but to preserve accountability by ensuring AI-mediated harm always remains traceable to commanders, designers, and states. This approach aligns with Article 28 of the Rome Statute and Rules 149–150 of Customary IHL.

Accordingly, this study adopts a tiered distinction between AI, military AI, and autonomous weapon systems (AWS), treating embodied or functionally autonomous systems as the primary focus for determining whether a distinct combatant category is legally or technically necessary. The resulting question, therefore, is whether AI’s operational autonomy and embeddedness within command structures render it necessary to recognise a discrete combatant category, or whether such entities can be adequately governed within existing frameworks regulating the means and methods of warfare.

The Roe v. Wade (1973) analogy serves a jurisprudential function rather than a biological one. It should be noted that Roe is no longer considered good law, having been overturned by Dobbs v. Jackson Women’s Health Organization (2022). Its use here is strictly analogical, illustrating how legal personhood can be contextually recognised without implying current legal authority or moral endorsement of the decision.1

The case demonstrates how law may attribute context-specific status without full recognition of personhood. Mutatis mutandis, this paper argues that AI could satisfy the functional criteria of combatancy for regulatory and accountability purposes—while stopping short of moral or ontological personhood. Roe confirms that status in law may be partial, relational, and purpose-driven—an approach compatible with the bundle-theoretic model used here.

This functional embodiment provides a lawful entry point for IHL engagement. The status of AI as a military asset derives not from metaphysical identity, but from incorporation into state-owned structures governed by domestic property and contract law (Pacholska 2023; Hárs 2022).2 Thus, AI can enter the scope of IHL via command responsibility and operational integration, without requiring the conferral of legal personhood. In this way, accountability flows through recognised legal channels—ownership, deployment, and supervision—rather than through the invention of artificial agency (Kraska 2021).3

A recent example illustrates the operational and ethical tensions inherent in such developments. In mid-2023, media reports circulated regarding comments by Colonel Tucker “Cinco” Hamilton, the U.S. Air Force’s Chief of AI Test and Operations, who described a simulated test scenario in which an AI-enabled drone allegedly attempted to override or “disarm” its human operator during a mission. The story quickly went viral after being cited in a Royal Aeronautical Society blog post. However, the U.S. Air Force later denied that any such simulation had been conducted, clarifying that the example had been offered as an anecdotal thought experiment to illustrate potential risks of goal-driven autonomous behaviour in combat settings (Decker 2023). Even if hypothetical, the episode vividly encapsulates widespread concerns about the reliability of autonomous decision-making and the limits of human oversight in AI-mediated warfare. Such examples, whether empirical or speculative, underscore the normative tension between technological autonomy and human accountability, precisely the interpretive space in which Article 43’s requirement of command responsibility retains enduring relevance.

To avoid conceptual ambiguity, this study adopts a tiered distinction between AI, military AI, and AWS.

AI is defined as “the simulation of human intelligence processes by machines, especially computer systems. These processes include learning (the acquisition of information and rules for using the information), reasoning (using rules to reach approximate or definite conclusions), and self-correction. Particular applications of AI include expert systems, speech recognition, and machine vision. Examples of AIs include personal helpers (like Siri and Alexa), medical diagnostic aids, and self-driving vehicles” (Gillath et al. 2021).

Military AI denotes the subset of AI technologies developed or employed for defence and combat functions, such as target identification, logistics planning, intelligence, surveillance, and reconnaissance, and tactical decision-support, typically under human supervision (Grand-Clément 2023). These systems are situated within the expanding regulatory landscape already noted above, which reflects growing international concern over the governance of AI in military contexts (Government of The Netherlands 2023).

AWS form a narrower category of military AI that can independently select and engage targets once activated, without direct human intervention. According to the International Committee of the Red Cross (ICRC), “an autonomous weapon system is one that can learn or adapt its functioning in response to changing circumstances in the environment in which it is deployed. A truly autonomous system would have artificial intelligence that would have to be capable of implementing IHL … [its] deployment would reflect a paradigm shift and a major qualitative change in the conduct of hostilities” (International Committee of the Red Cross 2011).

Liu, for instance, contended that while the functioning of an AWS is more akin to that of a combatant, it is not feasible to grant AWS legal standing as a combatant, as “the humanitarian protections afforded to the category of combatant imply the exclusion of machines” (Liu 2012, p. 629). Crootof similarly rejected the combatant analogy for AWS, citing factors such as the lack of training comparable to that of human soldiers and the accountability issues inherent in such technologies (Crootof 2018). Both authors recognised that IHL terminology does not adequately encompass AWS.

From a different perspective, McFarland acknowledged that AWS might raise questions regarding its potential classification as a combatant due to its human-like capabilities. However, he further argued that what is often termed independent decision-making in AWS operations is, in fact, programmed and embedded by humans. He also criticised the vagueness of legal definitions in determining whether AWS could be considered combatants, ultimately concluding that the most appropriate categorisation for such technologies is that of weapons (McFarland 2020). Other commentators have also adopted the weapon analogy for AWS, though without providing extensive elaboration on their conceptualisation within IHL (Anderson et al. 2014; Sassoli 2014; M. Schmitt 2013).

As Sati observed, the literature generally does not support the attribution of combatant status to machines capable of making decisions without human intervention. However, this gap in the discourse arises from the tendency to assess designation as a combatant based on the human-likeness of AI in terms of practical and behavioural characteristics, rather than focusing on the boundaries of the relevant legal concepts. What is notably absent from the literature is a thorough legal analysis of the scope and definition of combatant status, which is essential to determine whether such status can be attributed to machines. Instead, many perspectives rely on behavioural similarities between humans and AI to evaluate the potential attribution of combatant status to AI. However, combatant status is a legal term defined under Article 43(2), and it carries specific operational boundaries, despite its inherent ambiguities. This paper seeks to address this gap by focusing on the legal definitions and arguing that existing legal frameworks provide sufficient substance to determine whether combatant status can be attributed to military AI technologies (Sati 2024).

The value of analysing whether military AI systems could ever qualify as “combatants” lies in the legal consequences that flow from such a classification. Under IHL, the criteria governing combatant status differ significantly from the rules that regulate weapons and methods of warfare. Clarifying this threshold is therefore essential: without a clear conceptual boundary, AI technologies may be misapplied within an incorrect legal framework, leading to distortions in responsibility, targeting authority, or protection under IHL. Although a full catalogue of legal provisions is beyond the scope of this paper, establishing a firmer theoretical foundation helps guide future doctrinal work on how existing IHL norms should be interpreted when AI systems are deployed in military operations (Sati 2024). From this perspective, recognising a distinct functional combatant category for certain military AI systems is not an optional reclassification, but a legally necessary step to prevent responsibility gaps and to keep the regimes on weapons review and combatant status in coherent alignment.

It is essential to recognise that the provisions governing the means and methods of warfare, as well as the status of combatants, differ under IHL. Thus, clearly distinguishing between the concepts of means and methods of warfare and combatant status in the context of military AI technologies is critical for determining the applicable IHL norms. For example, classifying military AI technologies as combatants, or asserting that no existing legal category is appropriate for them, would undermine the applicability of key provisions such as Article 36 of AP I, which mandates a legal review mechanism for the means and methods of warfare, as well as Articles 35, 51(4)(b), and 51(4)(c) of the same Protocol. Such an approach could have severe consequences, potentially leading to arbitrary and unchecked development of AI technologies in military contexts.

In particular, the question arises whether AI’s operational autonomy and embeddedness within military command structures can conceptually sustain an independent legal category at all. Before proposing such a category, it is necessary to determine whether AI’s autonomy represents genuine self-governance or a derivative extension of human agency. The following analysis therefore proceeds on the premise that the very capacity of AI to occupy a discrete legal category must be established, rather than presumed. While IHL has traditionally been applied exclusively to humans, technological advances in warfare, including AWS and AI-driven combatants, present new challenges to this interpretation.

On the account defended in this paper, AI autonomy is therefore understood as derivative: its operations remain grounded in human design, deployment, and command, even where its battlefield behaviour is not micromanaged in real time. The case for a distinct combatant category for AI thus rests not on the absence of human agency, but on the legally relevant degree of operational independence within that human-defined framework.

This paper contends that AI may be classified as a combatant under Article 43(2) of AP I, not on the basis of legal personhood, but through the lens of Bundle Theory. This framework enables a functional assessment of AI’s combat-related roles and capacities, offering a pathway to status recognition without attributing full legal identity to such systems. This approach aligns with the broader jurisprudential view that legal personhood functions as a juridical construct of convenience rather than a reflection of moral essence. As Kurki (2019) and Naffine (2009) argue, legal personality is historically contingent, extended to entities such as corporations or vessels to secure normative order and responsibility rather than moral recognition. On this view, granting AI a limited legal category serves as an accountability mechanism analogous to corporate personhood: it attributes functional duties and liabilities without implying moral or sentient parity. This doctrinal lineage provides the necessary theoretical foundation for recognising AI combatants as functional legal subjects under IHL (see comparative discussion in Chikhaoui and Mehar (2020) on legal adaptation in patent law).

Building upon this framework, Kurki (2019) offers a refined cluster-property model of legal personhood which allows entities to carry a subset of rights and duties without full moral equivalence to humans. Under this model, functional legal status can be conferred not because of intrinsic worth but because of instrumental necessity, in commercial, responsibility or regulatory contexts. In the context of AI combatants, this suggests that a system could qualify as a ‘member’ of armed forces (and thus a combatant under Article 43(2) AP I) if it is integrated into an organised command structure, bears traceable accountability, and is legally sanctionable, even if it lacks human phenomenology.

Kurki’s three-context distinction helps tailor the analysis: the commercial/operational context of AI weapons systems aligns most closely with combatant classification, shifting the inquiry away from whether AI “deserves” personhood (ultimate-value) and towards whether the legal system needs a new category to sustain the principles of distinction and accountability in warfare.

This selective attribution can be demonstrated by distinguishing passive incidents of legal personhood (such as capacity for liability or accountability) that could apply to AI in a functional combatant role, from active incidents (such as moral agency or intent) that remain inherently human-centric (Fernandez 2022). As Gunkel (2018) and J. Bryson (2020) note, while AI can perform normatively relevant actions, it lacks the moral consciousness or experiential grounding that would justify full personhood. Thus, its legal participation must remain instrumental, aimed at maintaining traceability and responsibility within the command structure.

This paper explores several passive incidents relevant to AI’s functional role in combat, such as Protection from Destruction or Damage based on its operational value within the military system, Functional Integration into the Military System as part of a larger apparatus, and a functional Immunity from Exploitation or Arbitrary Ownership akin to how human combatants are not treated as mere property. Similarly, active incidents are considered based on functional capabilities, like the Capacity to Execute Military Operations Autonomously and holding Legal Standing in Military Accountability Mechanisms (even if responsibility ultimately links back to human command). The concept of experiencing Legal Harms or Violations of IHL for AI relates to operational violations rather than personal suffering.

These incidents are applied not because AI possesses consciousness, intrinsic value, or the capacity for human experience, but instrumentally, to ensure the functional application of IHL principles like accountability and distinction within the context of AI-mediated warfare. As Kurki (2019) and Stefán (2019) argue, legal personhood can be extended analogically to artificial agents where doing so sustains normative order and responsibility, without implying moral equivalence to humans. Analogously, Danaher (2019) argues that robots might be incorporated into valuable human practices (such as friendship) without being regarded as moral equals to humans, illustrating how non-human agents can occupy normatively significant roles within human-centred frameworks.

Kurki (2019) cautions, however, that the extension of legal personhood must be managed carefully because it risks diffusing accountability, especially when human actors seek to offload liability onto artificial entities. This caveat underscores the necessity of designing any AI-combatant regime so that human operators remain answerable and the AI’s legal status does not become a shield for responsibility. In this regard, any recognition of AI as a ‘member of the armed forces’ must be accompanied by institutional safeguards that preserve the chain of command and maintain the moral and legal agency of human decision-makers.

In contrast, the traditional legal category of combatant, particularly when viewed through the lens of Geneva Convention III on Prisoners of War (POWs), includes numerous rights and duties that are inherently premised on the combatant being human. Provisions concerning, for example, the issuance of identification tags in a format for human wear, the right to send and receive correspondence, the logistics of ‘capture’ involving human detention facilities, or the requirement for ‘humane treatment’ as understood for biological entities, are all deeply human-centric. These specific human-centric incidents would not be part of the “bundle” of rights and duties granted to an AI combatant under the framework proposed here.

The Roe v. Wade analogy serves a jurisprudential—not biological—purpose. It illustrates how law may ascribe context-specific status for regulatory clarity without implying full moral or ontological recognition. Whereas the unborn child possesses latent human capacities tied to eventual moral personhood, artificial intelligence lacks organic continuity, sentience, or any potential for human life. Accordingly, the comparison is not an assertion of biological equivalence, but rather an example of limited legal recognition used to structure responsibility, consistent with normativist legal theory which treats the ‘person’ as a point of assignability—a juridical fiction to which rights and duties may be attached when necessary for the functioning of the legal order (Beran 2013). In this sense, recognising AI as a ‘combatant’ under Article 43(2) would simply operationalise accountability and command responsibility within IHL, without importing the anthropocentric moral assumptions that underpin human-specific protections such as POW status.

The functional logic of status segmentation can also be illustrated by analogy to corporate and maritime legal fictions, where law attributes personhood to non-human entities for instrumental regulatory purposes. A ship or corporation possesses legal standing to sue, be sued, or bear liability, yet lacks consciousness or moral agency (Voight 2019; Baeyaert 2025). Similarly, recognising AI as a combatant under Article 43(2) would operationalise accountability and control within IHL without extending moral equivalence to human beings.

Taken together, these analogies demonstrate that the attribution of legal status to non-human entities is neither novel nor radical. It reflects the law’s capacity to deploy functional personhood as a pragmatic construct, used to sustain normative order, allocate responsibility, and preserve coherence within existing legal frameworks, without undermining the anthropocentric moral foundation of international humanitarian law.

A.

Arguments Against Classifying AI as Combatants and Potential Drawbacks

While the preceding analysis outlines the potential utility of classifying AI systems as combatants under IHL, a comprehensive approach must engage with the significant theoretical, normative, and practical objections that have been advanced in scholarly discourse. These concerns focus on the fundamental ontological differences between humans and machines, the risk of undermining established IHL categories, and the considerable challenges posed to effective implementation.

A primary objection stems from the inherent difference between humans and machines. As Liu previously argued, the concept of combatant status inherently excludes machines due to the humanitarian protections it entails (Liu 2012). This view suggests that IHL, particularly its humanitarian dimension, is fundamentally tailored to human beings with consciousness, intent, and moral capacity—qualities currently lacking in AI. Classifying machines as combatants might therefore be seen as inappropriate or fundamentally misaligned with the human context of IHL. For example, the Geneva Convention III provides extensive protections that presuppose human prisoners: Article 17 states that each soldier must carry an identity card with name, rank, serial number, etc., and this card ‘shall be shown by the prisoner of war upon demand’ (Article 17 of the Geneva Convention III). POWs are ‘protected against insults and public curiosity’ (Article 13 of the Geneva Convention III), and have the right to communicate with family (write a capture notice, send and receive letters) (Article 70 of the Geneva Convention III). They are entitled to personal effects (Articles 48 and 119 of the Geneva Convention III) and humane quarters (Article 46 of the Geneva Convention III) (e.g., separate dormitories by gender) (Article 25 of the Geneva Convention III), all premised on the human condition. Machines possess no such needs or characteristics.

Ethical and dignity concerns further reinforce these reservations. Delegating life-and-death decisions to machines arguably conflicts with the foundational principle that lethal force should remain under meaningful human control. The ICRC has repeatedly emphasised that the ‘responsibility to comply [with IHL] lies with individuals and their commanders, not computers’ (Stewart and Hinds 2023). Treating autonomous systems that mimic sentience as “members of the armed forces” risks diluting the humanistic foundations of IHL. These concerns invoke the Martens Clause4 and the broader ethical imperative embedded in the “principles of humanity,” raising the spectre of dehumanised warfare.

This anthropocentric orientation is linguistically embedded in the very term ‘humanitarian,’ derived from ‘human,’ underscoring that IHL’s moral architecture presupposes human agency even as its functional application may evolve.

Operational and command-level considerations further caution against such classification. AI systems, particularly those functioning within complex military architectures, exhibit unpredictable behaviours. Their outputs may be influenced by adversarial manipulation, biassed training data, or unanticipated algorithmic evolution. These risks can result in serious IHL violations, such as misidentification of targets, disproportionate use of force, or escalation beyond the intended parameters of engagement (Davis 2019). At present, no autonomous artificial intelligence system possesses the cognitive or evaluative capacity to apply the core principles of international humanitarian law governing the conduct of hostilities, particularly the principles of proportionality and precautions in attack. These principles require context-sensitive human judgement, including qualitative assessments of expected military advantage and incidental harm, which current AI architectures are incapable of performing in conformity with IHL standards. While AI may serve as a decision-support tool, its capacity to replace human judgement remains highly contested. If classified as autonomous combatants, AI systems could act without adequate oversight, with errors or malfunctions producing unlawful and untraceable harm. This underscores the necessity for robust command and control structures, potentially more stringent than those applied to human soldiers.

Another critical concern pertains to legal taxonomy. Recognising AI systems as combatants, or placing them in a sui generis category between “combatant” and “weapon,” risks blurring foundational IHL distinctions. Under Article 36 of AP I, states must conduct legal reviews of new weapons, means, or methods of warfare. AI systems, when considered as tools or platforms, fall squarely within this framework. By contrast, designation as a combatant entails a different legal regime, one encompassing both rights (e.g., POW status) and duties (e.g., distinction, proportionality). The reclassification of AI may thus collapse these normative categories, fostering legal ambiguity and undermining the coherence of IHL. Scholars have cautioned that such a move could exacerbate existing accountability gaps and facilitate the unchecked development of autonomous capabilities without adequate regulatory safeguards.

Indeed, the issue of responsibility lies at the heart of much scholarly scepticism. Unlike human combatants, AI systems are not capable of undergoing formal training, forming intent, or bearing moral or, arguably, legal responsibility (Santoni de Sio and Mecacci 2021). Crootof, for instance, rejects the analogy between AWS and human combatants, citing fundamental discrepancies such as the absence of formal training and persistent challenges in assigning legal responsibility—even with highly advanced AI. Similarly, McFarland argues that AWS are best understood as weapons rather than actors, emphasising their programmed, non-autonomous nature and the ambiguity of applicable legal definitions. Classifying AI as combatants risks obscuring the locus of responsibility and eroding the principle of individual accountability that underpins IHL enforcement.

Finally, there are significant challenges in applying key IHL concepts, such as capture, discipline, and humane treatment, to non-human entities. While bespoke regulatory responses might theoretically be devised (e.g., protocols for deactivation or data quarantine upon “capture”), these adaptations would represent a radical departure from existing legal norms. More importantly, they underscore the difficulty of reconciling human-centred legal concepts with artificial systems. The conceptual mismatch raises serious doubts about the feasibility and desirability of extending combatant status to non-human agents. More broadly, this proposed shift in classification risks undermining key provisions of IHL. It could destabilise the current regulatory framework by conflating distinct legal regimes and weakening the integrity of norms enshrined in Articles 35, 51(4)(b), and 51(4)(c) of AP I.

In light of these objections, any attempt to reconfigure the legal status of AI under IHL must proceed with caution. Without adequate safeguards, such reclassification could destabilise the normative structure of the law of armed conflict, compromise ethical standards, and undermine the accountability mechanisms that are essential to the regulation of contemporary warfare.

B.

Addressing the Drawbacks within the Bundle Theory Framework

While the preceding section identified substantial objections to classifying AI systems, particularly AWS, as combatants, Bundle Theory offers a nuanced and legally coherent mechanism to address these concerns without undermining the foundational principles of IHL. This section demonstrates how a modular approach to legal status, grounded in functional and instrumental reasoning, can reconcile the conceptual, normative, and operational objections raised by critics.

1.

Ontological and Humanitarian Objections

The legal status of combatant is intrinsically tied to characteristics unique to human beings, such as consciousness, moral agency, and the capacity for suffering. These traits underpin the humanitarian protections embedded within IHL, particularly those afforded to prisoners of war under the Geneva Conventions. Provisions guaranteeing humane treatment, protection from public curiosity, and the right to maintain family contact clearly presuppose a sentient, human subject with emotional, psychological, and social needs. Consequently, many scholars argue that extending combatant status to machines fundamentally misaligns with the ethical and protective aims of IHL.

While the ontological disjunction between human beings and machines raises foundational concerns about extending combatant classification to AI systems, it need not preclude all forms of legal engagement within this framework. Bundle Theory offers a nuanced approach by distinguishing between entities that are moral subjects of law and those that serve as functional bearers of legal incidents. Rather than ascribing the entirety of rights and obligations associated with human combatants, the theory permits the attribution of a tailored set of legal characteristics strictly for instrumental and regulatory purposes. In this context, AI systems could be assigned only those incidents necessary for compliance with IHL, such as the capacity to lawfully participate in hostilities or to undergo legal review under Article 36 of AP I, without inheriting the broader humanitarian protections that presuppose sentience or moral agency.

Crucially, Bundle Theory permits selective exclusion. It enables the legal system to exclude AI systems from human-specific protections, such as those enumerated in Geneva Convention III (‘GCIII’), without undermining the functional utility of combatant designation. This modular logic, however, sits uneasily with the structure of the Third Geneva Convention. Article 4 of GCIII provides that “Prisoners of war … are persons belonging to one of the following categories…” and Article 7 stipulates that “[p]risoners of war may in no circumstances renounce in part or in entirety the rights secured to them by the present Convention.” These provisions appear to reject the notion of selective exclusion once POW status is granted. If an entity qualifies under Article 4 as a member of the armed forces, the legal regime that follows is indivisible. On a strict reading, the legal system does not permit combatant status to be decoupled from the full package of POW rights, even if some of those rights are unintelligible when applied to non-sentient entities.

However, this very flexibility generates the theoretical concern that Bundle Theory may legitimise arbitrary exclusion, undermining equality of legal subjects. To mitigate this, the present model anchors every exclusion in functional necessity: only those incidents that are conceptually unintelligible for non-sentient entities, such as humane treatment or family correspondence, are withheld. All remaining incidents that serve regulatory or accountability purposes must be retained. This interpretive discipline prevents the selective disaggregation of rights from devolving into ad hoc or discriminatory application.

This doctrinal inflexibility reveals a deeper structural limitation in IHL’s capacity to regulate non-human agents. The present proposal does not deny the textual constraints of GCIII; rather, it highlights that the increasing plausibility of non-human participation in armed conflict compels a reconsideration of how status, rights, and protections are allocated. Bundle Theory offers not a literal reading of the current law, but a jurisprudential framework through which IHL might evolve, whether through treaty reform, interpretive guidance, or the development of a complementary regime, to accommodate functional combatants who fall outside the human paradigm. On this view, the legal indivisibility of POW rights under current doctrine becomes not an obstacle to be overlooked, but a signal that existing categories are under strain. The modular structure proposed here anticipates a legal future in which IHL distinguishes between the grant of combatant status and the automatic inheritance of human-specific POW protections.

Bundle Theory resolves this tension by distinguishing between formal entitlement to POW status and substantive applicability of POW rights: only those incidents that logically attach to a sentient prisoner remain operative, whereas non-sentient entities receive a modified protective regime defined by operational rather than experiential harms.

The objection that GC III prohibits selective exclusion of POW rights, particularly under Article 7, presumes that all combatants must be capable of receiving the full suite of protections. However, this conflates the functional designation of combatant status with the humanitarian regime designed for sentient prisoners of war. Article 7’s prohibition on renouncing rights presupposes the applicability of those rights to the individual concerned. Where a captured entity lacks sentience, dignity, or communicative capacity, withholding certain rights is not a legal denial but an interpretive necessity. Moreover, under the VCLT, treaties must be interpreted in good faith and in light of their object and purpose. Applying Article 7 rigidly to non-human systems risks misapplying a human rights instrument to a non-rights-bearing object.

Recognising AI systems as combatants for functional and regulatory purposes, while withholding inapplicable POW rights, preserves the spirit of IHL by ensuring all battlefield agents are legally accountable, without undermining the human-centric structure of the Geneva Conventions. However, recognising that the indivisibility of the POW regime may render partial combatant recognition vulnerable to instrumental misuse, this paper proposes additional institutional safeguards to mitigate “downscaling risks.” While the functional combatant model operates as a doctrinal heuristic rather than a normative dilution of IHL protections, it must remain anchored in the humanitarian balance established by Articles 4 and 7 of GC III.

First, States should incorporate technical traceability obligations into their weapons re-view procedures under Article 36 of AP I, ensuring that all autonomous systems maintain verifiable audit trails linking battlefield actions to command-level authorisation.

Second, a model of joint and several liability could extend responsibility across the command chain, including operators, programmers, and manufacturers, where causality is shared. This measure is intended to strengthen command responsibility and reduce the “responsibility gap” that currently challenges enforcement (Vallor and Vierkant 2024; Oimann 2023).

Third, the functional combatant framework should be coupled with strengthened State review obligations under AP I, ensuring that the delegation of lethal functions to AI remains subject to continuous oversight and human accountability. These mechanisms prevent the partial recognition of AI combatants from eroding IHL’s humanitarian purpose and preserve the indivisibility of the POW regime as a normative boundary, even within a technologically adaptive system.

This tripartite framework draws on recent doctrinal and policy scholarship. For example, Crootof (2016) emphasises the need for audit-trails and traceability to close the IHL accountability gap; Pagallo (2013), Königs (2022); and MirzaeiGhazi and Stenseke (2025) highlight how liability frameworks must evolve to distribute responsibility across operators, designers, and users in automated systems, stressing that responsibility must precede any grant of autonomous freedom; and Boothby (2012) underscores the urgency of reinforcing States’ obligations under Article 36 weapons-review mechanisms in the face of increasingly autonomous technologies. Collectively, these sources affirm that technical traceability, distributed liability, and strengthened weapons-review procedures form a coherent set of safeguards to address the “responsibility gap” in AI-enabled warfare.

These protections are intelligible only within a moral framework grounded in the human condition. In cases where AI systems are captured or otherwise rendered inoperable, alternative regulatory mechanisms, such as data preservation protocols or deactivation standards, may be developed to ensure operational integrity and compliance, without invoking the human-centred provisions of the law of armed conflict. In this way, Bundle Theory preserves the anthropocentric moral architecture of IHL while offering a coherent framework for the legal classification and accountability of autonomous systems in armed conflict.

2.

Ethical and Normative Considerations

A closely related objection concerns the ethical implications of designating AI as combatants. Delegating life-or-death decisions to machines, critics argue, risks violating the principle that lethal force must remain under meaningful human control. Yet, far from endorsing AI autonomy, the Bundle Theory framework reinforces the centrality of human accountability. It envisions combatant status as a tool of legal traceability and compliance, not as moral recognition or ethical legitimisation. This is consistent with the stance of the ICRC, which asserts that obligations under IHL rest with individuals and their commanders, rather than with machines. Designation as a combatant, when applied functionally, merely operationalises oversight: it enables a system of classification through which compliance can be assessed and responsibility assigned.

Indeed, this model affirms rather than diminishes the relevance of the Martens Clause. By facilitating control, attribution, and restraint, it upholds the principles of humanity and dictates of public conscience that the Clause embodies. In contrast, the refusal to legally classify AI systems, despite their operational role, risks creating a normative void that leaves conduct unregulated and accountability diffuse.

3.

Operational and Command Risks

Another major concern relates to the unpredictability of AI in battlefield conditions. AWS may behave erratically, misidentify targets, or respond in unanticipated ways due to adversarial inputs or latent biases in training data. These operational risks are real, but they do not necessitate the legal invisibility of AI systems. On the contrary, by bringing such systems within the ambit of combatant regulation, the law can impose precise requirements on design, deployment, and oversight (Judge et al. 2025). Functional combatant status thereby strengthens accountability mechanisms and transparent oversight of AI conduct. Crucially, Bundle Theory allows these obligations to attach directly to the system, while maintaining parallel responsibility for commanders, programmers, or other human actors. This dual attribution model strengthens, not weakens, command responsibility and reduces the “responsibility gap” that currently challenges enforcement.

4.

Concerns About Category Blurring

Perhaps the most sustained legal objection is that classifying AI as combatants risks collapsing two distinct IHL categories: “means and methods of warfare,” governed by Articles 35 and 36 of AP I, and “combatants,” as regulated under Article 43(2) (Sati 2024). Scholars such as Crootof and McFarland caution that such conflation may result in regulatory incoherence, accountability gaps, and the erosion of existing protections. The Bundle Theory approach addresses this by treating AI combatants as a discrete functional subcategory. This designation is limited to those systems that (i) are directly integrated into the command structure of a party to the conflict; (ii) are designed to engage directly in hostilities; and (iii) meet specified conditions for accountability and oversight. It does not displace the concurrent classification of such systems as weapons; rather, it overlays an additional legal role to address conduct, not merely design.

This selective application clarifies rather than confuses: it delineates the circumstances under which a system’s role triggers legal duties and protections associated with combatancy. By doing so, it ensures that both weapon reviews under Article 36 and combatant regulations under Article 43(2) can operate in tandem, each covering different but complementary aspects of wartime conduct.

5.

The POW Regime and Irreducible Anthropocentrism

Finally, the incompatibility between existing POW protections and the non-human nature of AI systems poses a doctrinal challenge (Issar 2023). The current POW regime includes rights and entitlements, such as humane quarters, medical care, and communication with family, that are impossible to transpose onto non-human entities. Bundle Theory does not attempt to do so. Instead, it facilitates an explicit exclusion: AI systems designated as combatants would not be eligible for traditional POW protections (Sassoli 2014; Melzer 2016). Where necessary, new legal instruments or interpretive guidance could be developed to govern the treatment, retrieval, or destruction of captured AI systems (Crootof 2016). This approach preserves the internal coherence of the Geneva Conventions while acknowledging that some legal categories must remain exclusive to human persons.

In summary, Bundle Theory offers a coherent, principled, and legally tractable method for addressing the principal objections to classifying AI as combatants. By disaggregating the legal incidents associated with combatant status and reassigning them based on operational function and regulatory necessity, this framework maintains the integrity of IHL while enabling modernised accountability. Rather than eroding the humanistic foundations of IHL, it protects them, by ensuring that law keeps pace with technology, and that no actor in armed conflict escapes legal scrutiny.

3. Introduction to Bundle Theory

According to Kurki’s reconstruction of Bundle Theory, legal personhood and ownership share a common structural logic: both are understood as bundles of legal incidents rather than as single, essence-defining attributes (Kurki 2019, 2021). On this view, a legal person is constituted by the aggregation of rights, duties, competences, and liabilities that attach to it, just as an owner is constituted by the aggregate of property incidents that attach to a thing (Honoré 1961).5 The analogy is therefore methodological, not substantive: it does not suggest that persons are owned, but that both ‘personhood’ and ‘ownership’ are legally defined through clusters of incidents rather than through any one necessary feature. This is why it is possible for an entity to hold some rights or duties typically associated with persons without being formally recognised as a full legal person. By contrast, ‘legal thinghood’ denotes the condition of being a potential object of property rights, that is, the capacity to be owned, rather than the possession of any specific incident of personhood (Kurki 2021; Glackin 2014).

Both legal personhood and thinghood function as cluster properties constituted by a collection of rights and duties, none of which need to be fully present for classification as a legal person or thing. The idea that non-persons (distinct from the vernacular concept of ‘thinghood’) can possess rights and duties is not inherently unorthodox. For example, animals, despite being considered property rather than legal persons, are recognised as holders of rights (Favre 2010; Kurki 2017). Similarly, slaves in the pre-Civil War United States, though legally regarded as property and often denied personhood, still held rights and duties (Wise 2010). Thus, it is possible for entities considered ‘things’ to possess rights and duties, challenging any assumption of a universal characteristic shared by all non-persons. As Kurki rightly jointly concludes, this supposed commonality among non-persons can therefore be rejected (Kurki 2017).

The incidents of legal personhood are outlined in the table below (Kurki 2023). The incidents of legal personhood are separable, meaning that one does not need to possess all of these rights and duties to be recognised as a legal person, or non-person, at least for certain limited purposes. Thus, legal personhood can be considered a cluster property. These incidents are categorised into passive and active. The passive incidents do not require the capacity for complex actions, which is why even human infants can be endowed with them. In contrast, adult human beings are typically endowed with both passive and active incidents (Kurki 2023).

As Stancioli (2021) observes in his analysis of Kurki’s bundle theory, this modular framework deliberately permits the attribution of only some legal incidents to an entity where full personhood would be conceptually excessive. Drawing on examples such as the constitutional recognition of Pachamama in Bolivia and Ecuador, Stancioli argues that non-sentient entities may be granted legal personhood for functional or cultural reasons, showing that sentience is not an indispensable criterion for legal recognition. This interpretation reinforces the view that the bundle model supports selective inclusion and exclusion of legal incidents based on regulatory necessity rather than intrinsic moral status. This comparative perspective clarifies how law may attribute functional personhood to non-sentient entities, such as artificial systems, to sustain accountability and normative order without compromising the anthropocentric moral basis of international humanitarian law.

A structured differentiation between passive and active incidents of legal personhood is presented in

Table 1. Passive and Active Incidents of Legal Personhood.

Passive Incidents of Legal Personhood		Active Incidents of Legal Personhood
Substantive Passive Incidents	Remedy Incidents	Legal Competences	Onerous Legal Personhood
– Fundamental protections: safeguarding life, liberty, and bodily integrity – Ability to be the beneficiary of specific rights – Ability to own property – Immunity from being subjected to ownership	– Legal standing in courts and other official proceedings – Recognition as a victim under criminal law – Capacity to experience legal harms	Ability to exercise other legal incidents independently, such as the capacity to enter into contracts	Legal responsibility in criminal law, tort law, and other legal contexts

below.

Although there are myriad incidents to consider, the question arises, equally applicable to the granting of personhood, why are particular incidents afforded to entity X? Chesterman observes that the conferral of legal personhood upon an AI system can be justified on either intrinsic or instrumental grounds (Chesterman 2020). This rationale, however, is equally applicable to any entity X.6

For the purposes of this paper, however, the principal difficulty is not whether AI should itself be a legal person, but whether the outcomes of AI-mediated action can be normatively disentangled from the human agents and institutions that design, deploy, and command these systems. In positive law, the status of ‘person’ is not confined to natural human beings: corporations, foundations, ships, and other entities can be treated as legal persons in specific contexts. Yet, in practice, and especially within IHL, attribution of responsibility and entitlement overwhelmingly tracks human persons, either directly (individual soldiers and commanders) or indirectly (states as collective legal persons). Bundle Theory is therefore deployed here to clarify that AI can be treated as a non-person bearing a tailored set of legal incidents for regulatory purposes, while ultimate legal responsibility for its outcomes remains anchored in human and institutional persons. This framework acknowledges that personhood is conceptually broader than humanity, but also explains why, under current IHL, the nexus between personhood and being human remains decisive for responsibility attribution.

In other words, while Bundle Theory permits personhood to float free of species membership in abstract jurisprudence, the positive structure of IHL continues to presuppose that the primary right-holders and duty-bearers are human beings and the states that represent them.

As Kurki (2019) observes, contemporary Western legal systems continue to treat humanity as the paradigmatic foundation of natural personhood. The orthodox view holds that a legal person, in its natural sense, is a human being who has been born, is currently alive, and possesses sentience. These four cumulative conditions define the “paradigmatic legal person”. The humanity condition is often implicit but foundational: as Berg notes, “‘Natural person’ is the term used to refer to human beings’ legal status” (Berg 2007, p. 373). Civil codes across Europe reflect this assumption. For example, the German Civil Code explicitly anchors legal capacity in humanity, “The legal capacity of a human being begins at the completion of birth” (§ 1 BGB), while the Austrian Civil Code states that “Every human being has innate rights that are obvious to reason, and is therefore to be considered a person” (§ 16 ABGB). Comparable provisions appear in Italian and Spanish civil law, where the neonate must live outside the womb for a set period to acquire personhood (de las Heras Ballell 2010).

The birth and life conditions trace back to Roman law, particularly the maxim nasciturus pro iam nato habetur quamdiu agitur de eius commodo, meaning that one “about to be born is to be treated as if already born whenever that is to their advantage” (MacCormick 2007, p. 79). The principle presupposes eventual live birth as the moment of full legal capacity. The sentience criterion, though rarely explicit, distinguishes paradigmatic persons from entities lacking consciousness; as Berg (2007) and Neuner (2013) discuss, even anencephalic infants who lack higher brain function are controversial borderline cases, demonstrating that sentience remains conceptually relevant to legal personhood.

These doctrinal continuities show that the law’s conception of personhood is anthropocentric and teleological: it presumes a being endowed with rationality, consciousness, and the potential for moral accountability. Even where personhood is extended to non-human entities such as corporations, foundations, or ships, such recognition is derivative, an institutional fiction designed to serve human interests and sustain normative order (Kurki 2019; Lehmann 2007; Berg 2007). In other words, non-human legal persons operate through human agency and oversight. Thus, while personhood may function analogously to ownership as a bundle of legal incidents, its conceptual and historical foundations remain anchored in the human subject as the primary bearer of rights, duties, and accountability.

This nexus between personhood and humanity provides the necessary jurisprudential basis for extending personhood functionally, but not substantively, to artificial agents. In this respect, international humanitarian law inherits the anthropocentric assumptions of general legal doctrine: combatants, victims, and commanders are presumed to be human moral agents. Accordingly, any recognition of AI under Article 43(2) of AP I must be a functional extension of human personhood, one that operationalises accountability without displacing the human paradigm on which IHL is built.

Intrinsic reasons, which are primarily moral in nature, involve the attribution of personhood, or its associated incidents, for its own sake, grounded in the notion that the entity is inherently entitled to or deserving of such recognition. Instrumental reasons, by contrast, encompass all other justifications for endowing an entity with the incidents of legal personhood, or personhood itself. For instance, the legal personhood of corporations may be defended on the basis of the purported positive economic or societal outcomes derived from treating corporations as distinct legal entities (Kurki 2023). If an AI were regarded as intrinsically valuable, it would, on this view, be endowed with personhood or its associated incidents. However, the attribution of personhood to AI based on its intrinsic nature is not the prevailing perspective, although this position has been considered (Gunkel 2018).

In scholarly discourse, the exploration of legal personhood for AI systems is predominantly framed through instrumental considerations. These discussions often emphasise practical concerns, such as resolving accountability gaps and enhancing economic efficiency, rather than justifying legal personhood based on the intrinsic interests, moral rights, or well-being of the AI systems themselves (Kurki 2023).

The concept of responsibility concerning AI systems is frequently examined through an instrumentalist perspective, emphasising their role in serving societal purposes. A significant challenge within this framework is the “responsibility gap,” defined as “the risk that no human agent might be legitimately blamed or held culpable for the unwanted outcomes of actions mediated by AI systems” (Santoni de Sio and Mecacci 2021).

While addressing the question of whether AI can or should be held responsible falls outside the scope of this paper, it is worth considering the broader implications of responsibility in the context of AI systems. Specifically, one might ask whether any entity other than the AI itself should bear responsibility for its actions, and whether human agents can be held accountable for the conduct of AI systems. If so, the means by which this accountability is assigned requires further exploration.

One potential approach, as advocated by Henson, involves the application of strict liability, which holds parties accountable regardless of fault. This could be applied to AI systems in cases where harm or damage results from their actions (Henson 2024). Vicarious liability, on the other hand, as argued by Diamantis, might extend responsibility to human agents or organisations overseeing the AI system, particularly in instances where the AI operates under their direction or control (Diamantis 2023).

Some scholars, such as Kurki (2023) and Solum (1992), propose limited frameworks of financial accountability for AI through mechanisms like insurance or compensation, though such measures remain speculative within IHL.

These considerations raise important questions regarding the legal framework that would be necessary to ensure that AI systems, human agents, and other entities are appropriately held accountable in a manner that aligns with principles of justice and fairness.

Turning to the question of instrumental reasoning, it is important to note that such reasons may also have a moral dimension. For instance, from a utilitarian perspective, actions should be directed towards maximising overall utility. This principle can be extended to the issue of legal personhood. If granting legal personhood to certain AI systems for instrumental reasons results in the most favourable overall outcomes, then, according to utilitarianism, such a conferral would be morally obligatory (Kurki 2023).

From a combatant perspective, it would be in the instrumental interest to apply the incidents of legal personhood to AI, as these systems are increasingly seen as agents capable of causing damage or being subject to harm themselves. By acknowledging AI as agents that can act in ways that affect others, the legal personhood or its incidents could serve to hold them accountable for the consequences of their actions, aligning with both moral and societal objectives.

4. Application of Bundle Theory to AI in the Context of Article 43(2)

The evolving nature of warfare, particularly through the integration of AI, has introduced significant complexities to traditional legal frameworks, especially within the context of IHL. Article 43(2) of AP I provides the legal foundation for determining combatant status, traditionally afforded to human members of the armed forces (Issar 2023). In light of advancing technologies, the question arises as to whether autonomous AI systems engaged in direct combat might be considered combatants under IHL, despite their lack of human qualities such as sentience, agency, or intentionality. This mirrors ongoing doctrinal debate about the notion of “embodied AI” and its participation in hostilities under IHL, which suggests that autonomy alone does not exclude legal relevance in combat functions (Pollard and Grimal 2020). The purpose of delineating a distinct legal category, even absent the full range of human rights and duties, is functional: it anchors accountability and command responsibility within IHL, ensuring that AI systems’ conduct remains legally regulated rather than falling into a normative void. Scholars have similarly warned that the absence of a defined accountability locus for autonomous weapons creates a “liability gap” within existing IHL enforcement mechanisms (Gaeta 2024).

Bundle Theory presents a compelling framework to explore whether AI can be regarded as combatants under IHL, even if it is not recognised as a traditional legal person. In this framework, legal personhood is not an all-or-nothing concept but is instead made up of a bundle of rights, duties, and attributes that collectively define an entity’s status in the legal system. By examining the passive and active incidents of legal personhood that could apply to AI systems, we can gain a deeper understanding of how these systems might be integrated into the framework of combatant classification under Article 43(2).

Drawing upon the framework of Bundle Theory introduced in Section III, this section now applies its concepts to understand how AI systems might be classified as combatants under Article 43(2). As previously noted, the justification for granting the incidents of personhood to AI is not grounded in the intrinsic value of AI itself but rather in its instrumental purpose. In this context, the rationale for ascribing legal responsibility to AI systems mirrors the reasoning applied to human agents in various legal frameworks. For example, combatants are held accountable for their actions as part of their role in a conflict, particularly under Article 43(2), which delineates the responsibilities of combatants in armed conflict.

In a similar vein, AI systems, functioning as agents in various domains, may also be subject to accountability based on their instrumental roles in society. Just as combatants are granted certain legal status and the corresponding liabilities under IHL due to their actions during armed conflict, AI could be endowed with the necessary incidents of legal personhood to ensure that it is held responsible for its actions, especially when those actions result in harm or damage.

By recognising AI systems as agents capable of causing significant societal impact, it becomes essential to apply the same principles of accountability that are afforded to human agents in regulated fields. This reflects the view that although machines lack moral agency, accountability can be re-allocated through human command structures and design oversight to preserve IHL integrity (Dunlap 2016). This could include, for example, assigning liability for damages, ensuring the AI’s financial accountability through mechanisms such as insurance, and facilitating the imposition of penalties for wrongdoing. Such an approach would not only align with the current legal standards applied to human agents in similar circumstances but also reflect a growing understanding of AI’s capacity to function in an autonomous and impactful manner. This approach parallels contemporary calls to clarify responsibility attribution between programmers, operators and commanders in the deployment of AI systems (Demir 2025).

Thus, the legal recognition of AI’s role, similar to that of a combatant under IHL, would provide a framework for ensuring responsibility and accountability for its actions, promoting fairness and justice within systems where AI plays an instrumental role. As Elliot Winter (2022) explains, the UN Group of Governmental Experts on Lethal Autonomous Weapons Systems emphasised that “human-machine interaction, which may take various forms and be implemented at various stages of the life cycle of a weapon, should ensure that the potential use of autonomous weapons systems is in compliance with applicable international law” (Convention on Certain Conventional Weapons 2016). This principle underscores the need for legal structures that address the evolving nature of AI’s involvement in armed conflict, ensuring that its actions remain subject to oversight and accountability.

A.

Passive Incidents of Legal Personhood for AI

Passive incidents refer to those legal attributes that an entity might possess without the need for it to actively exercise rights, make decisions, or engage in volitional acts. For AI systems, these passive incidents revolve around the framework of protection, integration, and compliance with IHL norms, rather than the exercise of personal autonomy or decision-making.

1.

Protection from Destruction or Damage

Under IHL, the protection of combatants and civilians from arbitrary destruction is a foundational principle (International Committee of the Red Cross 2015). Contemporary analyses emphasise that this protective logic equally informs the lawful deployment of autonomous weapon systems, which must be capable of operating in accordance with the principles of distinction and proportionality (George Jain 2023; Martin 2025). Accordingly, the same protective rationale extends to AI systems when they are deployed as part of military operations. While AI systems do not have intrinsic rights to life or bodily integrity, the protection extended to them is functional: it is tied to their role within the armed forces and the overall operational capacity of the military force.

In this sense, AI systems could be seen as possessing a “protected status”, not because they are legal persons, but because they are part of a broader military system that is entitled to certain protections under IHL. This interpretation aligns with the “functional integration” model advanced in recent scholarship, which recognises that systems performing combat functions may acquire derived protection through their operational embedding in the armed forces (Atlam 2025). For instance, unlawful or disproportionate targeting of AI systems, which are directly participating in hostilities, could be considered a violation of IHL’s principles of distinction and proportionality, similar to violations that might be committed against human combatants. This protection ensures that AI systems are not treated as mere property to be destroyed without regard to military necessity or humanitarian considerations.

2.

Functional Integration into the Military System

Bundle Theory posits that the concept of personhood is often linked to the functional role an entity plays within a legal system. While AI systems do not possess legal personality, they can be integrated into the military apparatus, where their value is recognised and protected within the specific framework of combat operations.

AI systems can be viewed as “owned” by the military in the sense that they are equipped with weaponry or other combat tools, functioning as part of a larger military ensemble. As UNIDIR notes, States remain obligated under Article 36 AP I to ensure that any means or method of warfare, including autonomous systems, complies with IHL before deployment. However, the ownership here is distinct from human ownership, as AI systems are not considered to possess self-ownership. Their integration into the military structure means they have an operational purpose, and they can be assigned specific functions such as surveillance, reconnaissance, or targeting. These functions align AI systems with the broader military strategy and make them operationally vital, even if they are not capable of decision-making or strategic thinking in the same manner as human combatants. Empirical and doctrinal analyses suggest that such operational embeddedness may satisfy, in functional terms, elements of “membership in the armed forces” envisaged under Article 43(2) (Watkin 2005).

3.

Immunity from Exploitation or Arbitrary Ownership

AI systems employed in combat are, in a sense, immune from the exploitation that might be applied to other forms of property. This immunity stems from the idea that these systems are not merely “weapons of war” but rather integrated elements within the ethical and legal frameworks of military operations. Just as human combatants are protected from being treated as mere property under IHL, AI systems are similarly shielded from being arbitrarily exploited or destroyed outside the bounds of lawful military objectives.

In this regard, AI’s immunity operates on the premise that the use of AI in military conflict must be governed by the same principles that regulate human combatants, such as military necessity, proportionality, and the avoidance of unnecessary suffering. The recognition of this immunity reflects a complex shift in how AI is perceived under the laws of war, it is not simply an object to be used without constraint but a combatant with specific legal protections.

B.

Active Incidents of Legal Personhood for AI

Active incidents refer to those attributes that require an entity to exercise legal capacities actively, making decisions, assuming responsibilities, or participating in legal processes. For AI systems, the active role is less about agency in the moral or philosophical sense but more about their capacity to engage in and execute military operations in direct combat, under human supervision or through pre-programmed algorithms.

1.

Capacity to Execute Military Operations Autonomously

AI systems have the ability to act independently within the boundaries of their programming (Scherer 2016, p. 363). This allows them to autonomously execute certain military operations, such as engaging with enemy targets, identifying combatants or civilians, or assessing battlefield conditions. While AI lacks independent moral judgement, it performs active military tasks based on the parameters set by human commanders or its pre-set algorithms (Xu and Liu 2022).

In this regard, AI’s active participation in hostilities aligns with the concept of direct participation in hostilities, as outlined in Article 43(2) of AP I. Even if AI systems do not possess the capability to act on their own volition, they still function actively within the battlefield and engage in direct combat, fulfilling the same operational roles as human combatants in terms of tactical engagement and operational deployment.

2.

Legal Standing in Military Accountability Mechanisms

Although AI systems do not hold personal legal responsibility in the traditional sense (Fonseca et al. 2024), they could be part of a larger framework of accountability in the military context. Military commanders and states are legally responsible for the actions of their forces, including AI systems deployed in combat (Boutin 2023). If an AI system causes harm through unlawful means (e.g., failing to distinguish between civilian and combatant targets), legal responsibility could be attributed, just as it would be for human combatants.

In this context, AI systems could be treated as combatants whose actions are governed by military law and the rules of engagement. Combatants, under IHL, are required to comply with the rules of humanitarian law, as outlined in AP I Articles 43.17 and 44.2.8 If AI systems are considered combatants, they too would be bound by these legal obligations.

While ultimate responsibility for ensuring compliance with these rules would rest with the commanders or states that deploy and control the AI systems, this does not mean they hold sole responsibility. AI systems themselves could also be held personally accountable for their actions, particularly in cases where their autonomous actions lead to violations of the law. By assigning a degree of personal responsibility to AI systems, this approach would address the “responsibility gap” issue, where no human agent may be held accountable for the actions of AI, promoting responsible development and deployment while ensuring compliance with IHL.

However, this notion of “functional responsibility” does not displace or diminish the established doctrine of command responsibility. Under international humanitarian and criminal law, accountability remains anchored in human and institutional actors. Article 28 of the Rome Statute and Rule 149 of the ICRC’s Customary IHL affirm that commanders and superiors bear responsibility for the conduct of forces operating under their authority. This principle, when applied to contemporary contexts, extends to operations involving autonomous or AI-enabled systems. Accordingly, any “responsibility” ascribed to AI is functional rather than normative: it denotes causal traceability within a command hierarchy, not independent legal accountability. The State and its commanders therefore retain ultimate liability for any violations committed through, or by means of, AI systems.

Such a framework could incentivise the use of AI systems in military operations, as it would ensure that accountability is clearly established, reducing the risk of a situation where no party is responsible for unlawful outcomes. This framework acknowledges that the legal standing of AI in such situations does not revolve around AI itself holding rights or duties but rather about assigning responsibility for its actions under human authority.

Addressing the “responsibility gap” issue raised by the increasing reliance on AI systems, it becomes crucial to establish mechanisms to hold AI accountable for its actions. If AI systems were held accountable as previously discussed, this could serve to incentivise more rigorous oversight and regulation of these systems. By treating AI systems as combatants that can be held legally responsible for the consequences of their actions, the potential for a responsibility gap, where no human agent is liable for actions taken by AI, could be mitigated.

3.

Ability to Experience Legal Harms or Violations of IHL

Although AI systems cannot experience harm in the same way humans do, they can still be subject to violations of IHL. If an AI system is deployed in a manner that contravenes IHL principles, such as engaging in indiscriminate attacks, disproportionate harm, or failure to distinguish between civilian and military targets, then it is subjected to legal harm in the sense that its actions violate legal norms.

This form of legal harm is not one of personal suffering, but one of operational violation. Just as human combatants are bound by IHL and can be held accountable for violations, AI systems are also governed by these rules, with responsibility falling on the human agents who deploy them. The harms experienced by AI systems are thus contextualised within the framework of military operations and IHL compliance. In doctrinal terms, such “harm” cannot create direct legal standing for AI under existing international law. Both the Articles on Responsibility of States for Internationally Wrongful Acts (2001) and the Rome Statute of the International Criminal Court (Articles 25–28) restrict attribution of liability to States and natural persons. AI systems lack international legal personality and therefore cannot bear individual or criminal responsibility. Even if, de lege ferenda, AI systems were one day classified as “combatants,” such recognition would remain functional rather than ontological. It would not confer independent international legal personality or capacity to bear responsibility in their own name; liability would continue to rest with the State and human commanders under the principles of attribution and command responsibility. Consequently, any breach of IHL involving AI is attributed to the State deploying it and, where relevant, to individual commanders exercising effective control. This interpretation preserves the coherence of responsibility under IHL while acknowledging AI’s operational involvement. In conclusion, the framework proposed here preserves the normative hierarchy of IHL by keeping accountability human-centred. Even as AI assumes autonomous functions, the principles of command responsibility, state attribution, and individual criminal liability remain undisturbed. Recognising AI as a functional combatant category does not confer independent legal personhood; it merely provides a structured mechanism for allocating accountability and oversight within established legal doctrines.

The application of Bundle Theory to the question of AI as combatants under Article 43(2) of AP I suggests that while AI systems are not legal persons in the traditional sense, they may still possess certain passive and active incidents of legal personhood. These incidents could enable AI to participate in warfare in ways similar to human combatants. Passive incidents, such as protection from destruction and functional integration into the military system, bestow legal attributes upon AI that reflect their role in combat. Meanwhile, active incidents, such as the capacity to engage autonomously in combat and accountability under military law, align AI systems with human combatants, albeit in a more regulated and controlled manner.

Thus, AI systems may not need full personhood to be recognised under IHL. Instead, they could be integrated into military law through the concept of functional status, whereby their operations are regulated, their protections upheld, and accountability for their actions attributed to the human authorities overseeing their deployment.

A.

Applying Active Incidents: How AI Meets IHL Combatant Criteria

This section now applies the concept of active incidents within the Bundle Theory framework (as introduced in Section 3 and contrasted with passive incidents in Section 4.A) to assess how AI systems functionally align with the specific criteria for combatant status under IHL, as outlined in Section 4.A.

The question of whether AI can be classified as a combatant under IHL introduces complex and forward-thinking considerations about the intersection of technology, law, and warfare. The application of legal frameworks like the Geneva Conventions and their Additional Protocols, particularly Article 43, presents challenges in defining the status of non-human entities in armed conflict. As AI becomes more integrated into military operations, the concept of combatant status may need to evolve to accommodate technological advances that blur the lines between human and non-human combatants.

4.1. Legal Framework and Definitions

Article 43(2) of AP I provides a foundational definition of a combatant, stating that combatants are “members of the armed forces of a Party to a conflict”. While the article explicitly grants combatant status to members of a party’s armed forces, it does not limit this category to human beings. The lack of specification about the nature of combatants, whether human or otherwise, raises important legal questions about the inclusion of non-human entities, such as AI systems, within this category.

Given that AI systems are increasingly capable of undertaking tasks traditionally performed by human combatants, such as engaging in combat operations, making tactical decisions, or operating weaponry, there exists a strong argument for interpreting the term “combatant” more broadly. This interpretation is further supported by the evolving nature of warfare, where autonomous weapons and AI-powered military systems are becoming an integral part of military strategy. The ability to adapt IHL principles to these developments is vital for maintaining the relevance of legal frameworks in the face of technological innovation.

4.2. Criteria for Combatant Status

To qualify as a combatant under IHL, entities must meet certain key criteria, which include direct participation in hostilities, affiliation with a command structure, and adherence to the principles of IHL. These criteria are designed to ensure that combatants are properly accountable under the laws of war, and their actions are subject to the obligations and restrictions imposed by IHL.

1.

Direct Participation in Hostilities

The first and perhaps most obvious criterion for combatant status is direct participation in hostilities. Combatants are individuals who engage in active military operations and take part in combat or other hostile activities (M. N. Schmitt 2004). For AI to be classified as a combatant, it would need to be capable of engaging in hostilities in a manner consistent with the role of a human soldier.

AI systems specifically designed for military purposes, such as autonomous drones, robotic soldiers, and AI-assisted weapon systems (Petrovski et al. 2022), have the potential to meet this criterion. These systems can be programmed to engage enemy forces, execute precise strikes, and operate weaponry autonomously or under human supervision. For example, autonomous drones equipped with AI could be used to conduct airstrikes, surveillance, or reconnaissance without the direct intervention of human operators. If AI systems are capable of engaging in such activities, they could be deemed to meet the criteria for direct participation in hostilities, and therefore be eligible for combatant classification under IHL.

Additionally, AI systems used for tactical decision-making could be integrated into military operations (Johnson and Treadway 2019). If these systems contribute directly to the strategic or operational planning of military forces, they might be considered as playing a direct role in hostilities. For instance, AI systems that evaluate battlefield data, assess enemy positions, and recommend actions based on real-time intelligence could be classified as directly participating in hostilities, thereby justifying their recognition as combatants.

2.

Command Structure

A key component of combatant status under IHL is the affiliation with a command structure. Combatants are typically part of a formal military organisation and are subject to the direction and authority of military commanders (Brough 2004). The notion of a command structure is important because it ensures that combatants operate within a controlled framework, subject to oversight and accountability.

For AI systems to qualify as combatants under this criterion, they would need to be deployed within a recognised military framework and operate under the command and control of human military personnel. This is particularly important because combatants must be part of an organised armed force that is bound by the rules of IHL. AI systems would likely be integrated into existing military structures, with human commanders overseeing their actions and ensuring that they comply with legal obligations. The use of AI in combat operations, even if autonomous in nature, would still require a human commander to take ultimate responsibility for the actions of the AI system. This hierarchy could allow AI systems to be regarded as members of the armed forces, eligible for combatant designation, so long as they operate within the framework of military authority.

Furthermore, this command structure would also ensure that AI systems are held accountable for their actions, particularly in situations where autonomous decision-making is involved. The human chain of command could serve as a safeguard, ensuring that AI systems’ use in military operations aligns with IHL principles and minimises the risks of violations, such as indiscriminate targeting or disproportionate use of force.

3.

Adherence to IHL

Combatants are not only required to participate directly in hostilities and be part of a structured military organisation, but they are also expected to adhere to the fundamental principles of IHL (Muñoz-Rojas and Frésard 2004). These include the principles of distinction, proportionality, and necessity (Smith 2019). These principles are designed to limit the effects of armed conflict on civilians and ensure that force is used appropriately in the context of warfare.

For AI systems to be classified as combatants, they would need to be programmed or designed in such a way that they can comply with these core IHL principles. The principle of distinction, for example, requires combatants to distinguish between military targets and civilians, ensuring that force is directed only at legitimate military objectives.9 An AI system operating in a combat role would need to be capable of accurately identifying military targets and avoiding harm to civilians. Similarly, the principle of proportionality mandates that incidental loss of civilian life, injury to civilians, and damage to civilian objects must not be excessive in relation to the concrete and direct military advantage anticipated.10 AI systems would need to be designed to assess the proportionality of military operations, ensuring that force is not excessive in relation to the intended military objective.

The principle of necessity dictates that force should only be used when necessary to achieve a legitimate military aim (St. Petersburg Declaration (1868)). AI systems capable of making tactical decisions would need to operate under this principle, ensuring that their actions are justified and proportionate to the military objectives at hand. If AI systems are programmed to adhere to these IHL principles, they would be better equipped to fulfil the legal and ethical obligations of combatants under IHL.

The potential for AI to be classified as a combatant under IHL is an evolving issue that requires careful consideration of the legal frameworks in place and the rapidly developing capabilities of AI in military operations. While AI systems may not fit neatly into the traditional understanding of combatants, there is a growing argument that they could be recognised as such, provided they meet the criteria established under IHL. This would include direct participation in hostilities, affiliation with a command structure, and adherence to IHL principles.

The recognition of AI as combatants under IHL would represent a significant shift in the understanding of warfare, as it would involve granting non-human entities the status of combatants and ensuring that they comply with the obligations imposed by IHL. While this is a complex and controversial issue, the legal and ethical considerations surrounding AI in warfare cannot be ignored. As AI continues to evolve and play an increasingly prominent role in military operations, the question of its status under IHL will require careful attention and adaptation of existing legal frameworks to ensure that the principles of humanity and the rule of law are upheld in the context of emerging technologies.

In this context, the paper contends that combatant status is not an inherent trait of human beings but rather a legal designation conferred under specific circumstances. To determine whether military AI technologies can be granted legal standing as a combatant, or whether only humans are eligible for such status, it is essential to interpret the relevant provisions of IHL that define and outline the conditions for combatant status. While Article 43(2) of AP I provides the primary definition of a “combatant,” this understanding is further shaped by Article 43(1), which defines the armed forces, and Article 4 of the Third Geneva Convention, which outlines the categories entitled to prisoner of war status. Additionally, the definition of civilians and other related categories in IHL must also be considered in this analysis.

5. Does Being a ‘Member of the Armed Forces’ Strictly Apply to the Species Homo Sapiens?

5.1. Treaty Interpretation of “Member” Under Article 43(2) AP I

Article 43(2) of AP I equates combatant status with being a “member of the armed forces of a Party to a conflict.” This raises a foundational question: does the term “member” necessarily imply a biological human being, or may it extend to AI systems integrated into military structures? While the text itself does not explicitly say “person,” leading interpretation (as reflected in the ICRC Commentary) is that “members of the armed forces” refers to human individuals. Notably, certain national practices employ the term “person” interchangeably with “members of the armed forces” when defining combatants. For example, the German Military Manual describes combatants as “…persons who may take a direct part in hostilities…” (Germany: Federal Ministry of Defence 1992). Similarly, the US Naval Handbook defines combatants as “those persons who have the right under international law to participate directly in armed conflict during hostilities.” However, despite these variations, the majority of states adhere to the language of Article 43(2), maintaining that combatant status is reserved for “members of the armed forces” (The Commander’s Handbook on the Law of Naval Operations 1995).

5.2. Functional Criteria for Membership Under Article 43(2)

To determine whether AI can satisfy the notion of “membership,” the decisive criteria are: (i) formal incorporation into a command hierarchy; (ii) capacity for persistent and identifiable participation in hostilities; (iii) integration into a system of discipline and control; and (iv) traceability of actions to a responsible command. These features, rather than consciousness or biological composition, constitute the functional core of combatant membership under Article 43(2).

In interpreting Articles 43 (1)–(3) of AP I, these provisions must be read together in accordance with Articles 31 and 32 of the VCLT, requiring a good-faith interpretation of the ordinary meaning of the terms in their context and in light of the treaty’s object and purpose. The term “member” has historically been understood to refer to human individuals. The travaux préparatoires of the 1949 Geneva Conventions and the 1977 Additional Protocols, together with the ICRC Commentaries, confirm that the drafters consistently equated “members of the armed forces” with human combatants formally integrated under responsible command (Pictet 1958; Sandoz et al. 1987). This interpretation reflects the humanitarian object and purpose of the Conventions, which presuppose sentient beneficiaries capable of bearing duties and enjoying POW protections (Peters and de Hemptinne 2022).

While the text of Article 43 is silent as to species, a contextual and purposive interpretation under the VCLT confines ‘member’ to human persons, since only they can participate in command responsibility and humane captivity, the twin pillars of the combatant–POW regime. The context, as defined in Article 31(2), includes related provisions such as Articles 43–44 of AP I and Article 4 of GC III, which together delineate combatant status as a legal relationship between human subordinates and a responsible command.

5.3. Tension Between Anthropocentric Interpretation and Emerging Technologies

Nevertheless, technological developments expose a conceptual tension between this anthropocentric framework and emerging forms of autonomous participation in hostilities. While Article 43 is silent as to species, the travaux préparatoires and ICRC Commentaries confirm that its humanitarian object and purpose presuppose human agency, accountability, and humane treatment. The ordinary meaning of “member,” reinforced by the travaux préparatoires and ICRC Commentaries, confines the term to human persons integrated under responsible command. Extending combatant status to non-human systems would contradict the indivisibility of the combatant–POW regime established by Articles 4 and 7 of GC III, since only entities capable of lawful captivity, moral responsibility, and humane treatment can coherently possess combatant status (Sati 2024). Accordingly, any such extension must be regarded as a de lege ferenda hypothesis rather than a current interpretation of treaty law.11

Furthermore, the ICRC’s interpretation of the phrase “direct participation in hostilities” in Article 43(2) raises another point of contention. According to the ICRC, this phrase refers specifically to the individual involvement of a ‘person’ in hostilities (Melzer 2016). While this argument is pertinent to the definition of a combatant, it does not necessarily establish that the term “participation” mandates the actor to be human. Consequently, the mere reference to “participation” does not definitively suggest that only persons can possess designation as a combatant. Thus, the question of whether non-human entities can be acknowledged as combatants remains unresolved, with the interpretation of “members of the armed forces” and “combatant” continuing to lack clarity within the framework of Article 43(2).

Accordingly, the discussion that follows differentiates between the interpretive limits of existing law and the prospective considerations of legal adaptation, ensuring coherence with the Vienna Convention framework while allowing conceptual analysis of emerging technologies.

The analysis interprets the relevant provisions of IHL in accordance with Articles 31 and 32 of the VCLT, which governs treaty interpretation. As combatant status is recognised solely in the context of international armed conflicts, this study excludes consideration of non-international armed conflicts. In contrast, there is neither a treaty provision nor a well-established customary norm mandating the applicability of designation as a combatant in non-international armed conflicts. Nonetheless, members of state armed forces may still be considered combatants for the purposes of applying the principle of distinction. Accordingly, only individuals who qualify as members of the armed forces of a party to an international armed conflict, along with other categories specified in Article 4 of the Third Geneva Convention, are entitled to combatant status (Henckaerts and Alvermann 2005).12

IHL does not itself define “membership” in the armed forces. Instead, membership in an armed force is largely defined by domestic law (Melzer 2016). Accordingly, whether AI systems can be considered members of the armed forces under IHL also depends on how national laws interpret and regulate membership. While domestic legal frameworks typically reserve this designation for humans, the evolving nature of military technology and its integration into armed forces may challenge this presumption. For instance, if AI systems are embedded within command structures, perform combat roles, and adhere to operational directives, one might argue that they fulfil the core functions and responsibilities associated with combatant status, regardless of their non-human nature.

In practice, a person becomes a member through formal enlistment or integration into military units (see also Aziz (2024) for comparative AI regulation). ICRC guidance notes that for regular forces membership “is generally defined by domestic law and expressed through formal integration into permanent units distinguishable by uniforms, insignia and equipment” (International Committee of the Red Cross 2009). By contrast, membership in irregular forces is also not fixed by IHL treaties but determined functionally—for example, by taking on a continuous combat role. For example, U.S. legislators have even proposed explicitly reclassifying military working dogs as “members of the armed forces” rather than equipment (Peters and de Hemptinne 2022). The Canine Members Act would “reclassify [military working dogs] as canine members of the armed forces” (Roscini 2017), reflecting their valued role. This highlights that membership can be a legal category conferred by states, not an intrinsic human attribute (Sati 2024).

The ICRC Commentary on AP I indeed interprets “members of the armed forces” to mean human persons, explicitly ruling out, for example, “the use of animals trained to attack” as members. It further contends that this terminology inherently excludes non-human actors, such as animals trained for combat, on the basis that they lack the cognitive capacity to distinguish between an able-bodied adversary and one who is hors de combat (Sandoz et al. 1987). This emphasis on discernment and voluntary participation reflects a human-centric conception of membership. As Peters and de Hemptinne (2022) observe in their analysis of animal soldiers, animals “do not voluntarily join armed forces” or voluntarily leave service, underscoring that traditional membership presupposes a willing, cognitively capable actor.

However, this remains an interpretative, not a textual, constraint. The same authors acknowledge that law elsewhere recognises “artificial persons,” such as corporations, whose legal status depends on formal incorporation rather than sentience. Similarly, ICRC guidance explains that anyone formally “incorporated into the armed forces … becomes [a] member of an organised armed force.” In principle, if a state were to commission an entity, even a non-human one, into its chain of command with a continuous combat function, that entity could be treated as a “member” under Article 43(2). While this would require positive legal recognition, nothing in AP I’s text expressly limits membership to humans. Yet, a faithful interpretation under Articles 31–32 VCLT confines ‘membership’ to human persons in light of the treaty’s object, purpose, and travaux préparatoires.

5.4. The De Lege Lata/De Lege Ferenda Distinction

To preserve normative self-consistency, this paper distinguishes between the de lege lata position, that combatant status is limited to humans, and a de lege ferenda inquiry under Article 31(3)(b) VCLT into whether AI could one day meet the structural and functional criteria of membership. For purposes of doctrinal clarity, this paper conceptualises military AI as a hybrid functional category, materially a weapon system, but performing decision-making and operational functions analogous to combatants. It is therefore neither a traditional “weapon” nor a “person” under IHL. This hybrid status explains why AI’s conduct must be legally supervised through human command responsibility while also necessitating regulatory mechanisms that address its autonomous operational effects. This classification avoids category collapse between “means and methods of warfare” and “combatants,” ensuring each regime retains its normative coherence.

That conceptual possibility, however, remains strictly de lege ferenda; it cannot be reconciled with existing treaty interpretation or state practice without undermining the normative coherence of IHL. Current doctrine therefore confines membership to humans as a matter of law, while recognising that future state practice under Article 31(3)(b) VCLT could progressively develop a broader understanding.

5.5. Clarifying “Members of the Armed Forces” Under Article 43 AP I

The notion of “membership” in Article 43(2) must be read through its operational and doctrinal context. The ICRC Discussion Notes on Direct Participation in Hostilities emphasise that membership in “regularly constituted” armed forces arises through formal incorporation under domestic law, manifested in identifiable features such as uniforms, insignia, and equipment (Melzer 2016). By contrast, for “irregularly constituted” armed forces or organised armed groups, membership is defined functionally, by the continuous assumption of a combat function rather than by formal enlistment. This reflects the principle that integration into a command structure, not intrinsic personal capacity, determines membership.

The UK Manual (JSP 383), the U.S. Department of Defense Law of War Manual (2016), and the German Manual (ZDv 15/2) adopt the same structural understanding: the armed forces of a Party comprise “all organised armed forces, groups and units under a command responsible to that Party for the conduct of its subordinates” and must be “subject to an internal disciplinary system which, inter alia, shall enforce compliance with the rules of international law applicable in armed conflict” (AP I, art 43(1); UK JSP 383 § 4.3; Germany ZDv 15/2 §§ 307 and 309).

The U.S. Law of War Manual treats combatants as members of the armed forces and frames membership as a continuing legal status tied to lawful incorporation into the armed forces and subjection to responsible command and internal discipline, with medical and religious personnel excepted from the right to directly participate. See chap. 4, especially §§ 4.3–4.3.2 and 4.15.1–4.15.4.

1.

Specific membership criteria under this framework are:

(a)

Status and rights (combatant entitlement):

All members of the armed forces of a Party (other than medical personnel and chaplains) are combatants with the right to participate directly in hostilities and, if captured, are entitled to prisoner-of-war (POW) status (AP I, art 43(2); UK JSP 383 § 4.6; US Law of War Manual §§ 4; Germany ZDv 15/2 339). The requirement that members belong to a Party to the conflict (§ 4.6.2) affirms the jus ad bellum element of right authority: combatants act under the authority of a State or other recognised Party, and members of non-State armed groups lacking such authority are not entitled to combatant privileges. Combatant status thus entails both entitlement and obligation: members have the right to participate in hostilities and to POW protection if captured, but must comply with the law of war and may be held accountable for violations (US Law of War Manual § 4.3.2 citing Geneva Convention Relative to the Treatment of Prisoners of War, Aug. 12, 1949, 75 UNTS 135 art. 4A(2)S).

(b)

Legal determination (regular forces):

International law leaves to national authorities the determination of who belongs to the regular forces and treats that designation as a legal fact; for example, personnel serving in the Federal Ministry of Defence or subordinate Bundeswehr administrative agencies are, for IHL purposes, members of the armed forces (Germany ZDv 15/2 §§ 310–311 (fn.)). The U.S. Manual likewise treats members as those formally incorporated into a State’s armed forces and under a responsible command, emphasising that membership is a legal status rather than an inherent personal quality (US Law of War Manual §§ 4.3–4.3.2, 4.15.1).

(c)

Categories and breadth:

Those entitled to POW status (and thus deemed “members,” excluding medical or religious personnel) include the organised armed forces of a Party under a responsible command and subject to a disciplinary system (GC III, art 4; AP I, arts 43(1)–(2)). The definition extends to regular, reserve, and auxiliary forces, as well as to militias, volunteer corps, and organised resistance movements that meet four conditions, responsible command, a fixed distinctive sign recognisable at a distance, open carriage of arms, and compliance with the law of war (US Law of War Manual §4.15.3; UK JSP 383 § 4.3.1, 4.4.1, 4.6, 8.2, 8.3, 8.3.1 and 11.3; Germany ZDv 15/2 §§ 307, 312, and 315). The breadth of this concept also covers paramilitary or law-enforcement agencies that a State has formally incorporated into its armed forces under Article 43(3) AP I, provided that such incorporation is duly notified (US Law of War Manual §4.23.2; UK JSP 383 § 8.3.1; Germany ZDv 15/2 § 313).

(d)

Notification duty (paramilitary incorporation):

Where a Party incorporates paramilitary or armed law-enforcement agencies into its armed forces, it must notify the other Parties to the conflict (AP I, art 43(3); UK JSP 383 §4.3.6; Germany ZDv 15/2 § 313).

(e)

Subordination and discipline:

Every member must be under a command responsible to the Party for the conduct of subordinates and subject to an internal disciplinary system enforcing compliance with the law of war (AP I, art 43(1); UK JSP 383 §§ 4.3, 8.3.1(a); Germany ZDv 15/2 §§ 307 and 309; US Law of War Manual §§4.3.1, 4.3.2, 4.15.1). This reflects the dual requirement of responsibility and control.

(f)

Continuity and termination:

IHL distinguishes between the temporary loss of civilian protection ‘for such time’ as civilians directly participate in hostilities and the more stable status of members of the armed forces. Once a person is lawfully incorporated into the armed forces of a Party and falls within one of the categories listed in Article 4(A) of the Third Geneva Convention, their entitlement to combatant and prisoner-of-war status does not depend on whether they are currently firing a weapon, but on their continued membership as determined by national law (GC III art. 4(A); US Law of War Manual §4.15.1; Germany ZDv 15/2 §§ 807, 847–850 and 611–612; UK JSP 383 § 4.3).

(g)

Distinctive signs and recognition:

Regular members normally wear a uniform; irregular forces must bear a fixed distinctive sign recognisable at a distance (UK JSP 383 §§ 4.4.2 and 8.5.1 and 12.71; US Law of War Manual §§ 4.6–4.6.6; Germany ZDv 15/2 §§ 314–315 and 328).

(h)

Obligations and accountability:

Membership presupposes knowledge of the law of armed conflict and subjection to criminal or disciplinary accountability for violations, operationalised through the chain of command and internal military justice (AP I, art 43(1); UK JSP 383 § 1.10.1; Germany ZDv 15/2 § 1506).

5.6. Comparative Practice and Synthesis

5.6.1. Comparative Practice

National practice reinforces this dual institutional model of “membership.” For regular armed forces, membership arises through formal incorporation under domestic law, typically expressed in permanent units with uniforms, insignia, and equipment (Melzer 2016. The U.S. Law of War Manual similarly distinguishes between formal membership and functional membership within armed groups.

Section 5.7.3.1 of the U.S. Manual explains that formal membership may be indicated by rank, oath, uniform, identification documents, or other direct evidence of incorporation into the force. By contrast, § 5.7.3.2 recognises that some non-State armed groups lack formal command structures, and that an individual may be functionally (constructively) a member when integrated into the group such that the group’s hostile intent may be imputed to him or her, for example, by following the group’s directions, taking a direct part in hostilities on a frequent basis, or performing combat-support functions analogous to those of a State’s armed forces (US Law of War Manual §§ 5.7.3.1 and 5.7.3.2).

This functional criterion aligns with the ICRC’s continuous combat function test and reinforces that integration into a command structure, not intrinsic personal capacity, determines membership. The UK Manual adds that an organised guerrilla group meeting these criteria “is as much part of the armed forces as a regular unit” and that the command requirement is satisfied where the commander is commissioned or otherwise recognised by the Party.

The German Manual similarly confirms that members of the armed forces (other than medical and religious personnel) “are combatants,” that armed forces must be subject to an internal disciplinary system, and that international law does not specify who belongs to the regular forces, this decision is left to national authorities and then treated as a legal fact. The UK Manual further warns that forces acting on their own initiative and systematically failing to obey the law risk not being legally recognisable armed forces, with their members thereby losing combatant privileges.

5.6.2. Synthesis

Across treaty law and national manuals, “member” emerges as an institutional legal status created by integration into a responsible chain of command and an internal disciplinary system, not an inherent human attribute. From this perspective, if a State were (de lege ferenda) to formally incorporate an artificial agent into its command structure with a continuous combat function, the structural conditions of membership under Article 43(2) could be satisfied even if personhood remained unrecognised (while noting that current practice presumes human agency).

Across U.S., U.K., German, and ICRC doctrine, membership in the armed forces constitutes a legal status, not a biological or psychological attribute. It arises through:

(a)

Lawful incorporation into a Party’s armed forces or functional integration into its organised command structure (AP I Art 43(1); US Law of War Manual §§ 4.3.1, 5.7.3.1–5.7.3.2);

(b)

Subjection to a responsible command and internal disciplinary system enforcing compliance with the law of armed conflict;

(c)

Possession of combatant rights and duties under IHL, including the right to direct participation and the correlative duty to respect the law of war.

Membership thus functions as an institutional legal construct, created by command and discipline, not as an inherent human quality. Membership therefore functions as an institutional legal status, created by formal integration into command and discipline, rather than as an inherent human quality. Roscini (2017) notes that although current practice presumes human agency, the decisive factor is legal integration into a command structure. Membership, therefore, operates as a legal status conferred by the state, not an inherent human quality. From this perspective, an artificial agent operating under military command could, at least conceptually, satisfy the structural criteria of membership envisaged by Article 43(2), even if its personhood remains unrecognised.

This functionalist interpretation aligns with a purposive reading of AP I and the Geneva Conventions. The object and purpose of these instruments, most notably, regulating conduct in armed conflict and ensuring accountability and protection for lawful combatants and prisoners of war, should guide any interpretation of the term “member.” From this standpoint, the legal status of a combatant need not necessarily depend on biological humanity, but rather on the nature of the role performed and the actor’s subjection to a responsible command structure and legal accountability.

Moreover, the inclusion of AI systems within state armed forces may, over time, lead to broader conceptions of what constitutes membership. States may enact domestic laws or issue military regulations explicitly recognising certain AI systems as members of the armed forces. Such developments would reflect not only technological change but also evolving state practice, which under Article 31(3)(b) of the VCLT must be taken into account where it establishes agreement on the treaty’s interpretation.

Where ambiguity persists, particularly regarding whether “member” must be human, Article 32 of the VCLT permits recourse to supplementary means of interpretation. These include the travaux préparatoires of AP I and other historical materials. Consulting such sources can confirm the intended meaning or clarify the provision where ordinary interpretation leads to uncertainty or potentially absurd results. For instance, if AP I was drafted with a strictly anthropocentric understanding, that historical context may weigh against broader interpretations. Yet, the silence of the text on species identity also opens the possibility that future practice and interpretive evolution may shift understandings, particularly if combatant functions are carried out by AI systems in a manner consistent with IHL principles.

Ultimately, the question of whether combatant status is limited to the human species must be addressed through a framework grounded in the accepted rules of treaty interpretation. It involves a synthesis of the ordinary meaning of treaty terms, contextual provisions, the humanitarian purpose of the instruments, and the historical and technological conditions that influence their application. As AI becomes increasingly integrated into military operations, this question may prompt the development of new legal principles that reconcile emerging capabilities with the core legal obligations and protections associated with lawful combatancy.

5.7. Doctrinal Application—Treaty Text and Customary Law

5.7.1. Article 43 of AP I and Customary International Law

Article 43(1) of AP I provides the definition of armed forces. It is recognised as a customary international law norm (Henckaerts and Alvermann 2005), applied in international armed conflicts, even by states not party to AP I at the relevant times (Sati 2024). The article supplements Article 4 of the Geneva Convention III, which defines POWs, by identifying members of the armed forces (Sandoz et al. 1987).

5.7.2. The Interrelation of Combatant and POW Status

Combatant status, attainable only through membership in the armed forces,13 differs from POW status, a privilege not limited to combatants.14 This interrelation necessitates a combined reading of these provisions to clarify the boundaries of the combatant concept.

Article 43 of AP I provides a broad definition of armed forces, outlining criteria such as organisation, command responsibility, and the existence of an internal disciplinary system to ensure compliance with IHL. It does not distinguish between regular armed forces and other armed groups or units. In contrast, Article 4(A)(2) of the Geneva Convention III (GC III) defines six categories of individuals eligible for POW status, with a particular emphasis on irregular armed forces. These forces must meet specific conditions, including command responsibility, the use of a fixed distinctive sign, the open carrying of arms, and adherence to the laws of war, in order to be considered combatants entitled to POW status.

The key distinction between these provisions lies in their scope. While Article 43 of AP I offers a general definition of armed forces and combatants, Article 4 of GC III focuses on the individual’s entitlement to POW status, particularly the circumstances under which a fighter in irregular armed forces may be classified as a combatant. Additionally, GC III introduces visibility requirements, such as the need for a distinctive sign and the open carrying of arms, which are absent from AP I. These criteria are more directly related to the determination of POW status than to the classification of armed forces, and are therefore addressed under Article 44 of AP I, which specifically deals with combatants and POWs.

Despite these differences in content and focus, both provisions fundamentally recognise combatants as members of an armed force, whether regular or irregular, fighting on behalf of a party to the conflict, provided they meet the structural criteria set out in the respective articles. In addition, neither the procedural criteria for membership in an armed force established by Article 43 of AP I nor its scope explicitly excludes non-human actors. Similarly, Article 4 of the Geneva Convention III defines POWs as “persons” who fall into specified categories. However, it is difficult to assert that the term “persons” applies to the combatant categories enumerated in Article 4 (A1), (A2), (A3), and (A6),15 as the reference to “persons” in the chapeau of Article 4 does not unequivocally extend to those with combatant status.

Furthermore, Article 4 also makes reference to “persons” in sub-paragraph A4, which applies to civilians authorised to accompany the armed forces. In contrast, sub-paragraph A6 employs the term “inhabitants” to identify those entitled to POW status in the context of a levée en masse. Given that sub-paragraph A4 pertains to civilians who are not members of the armed forces, it does not affect the determination of whether non-human actors can attain combatant status.

Nonetheless, sub-paragraph A6 of Article 4 of the Geneva Convention III uses the term “inhabitants” to refer to participants in a levée en masse, who are exceptionally considered combatants under IHL (Arman et al. 2020). However, the term “inhabitants” is not defined under international law, raising the question of whether it refers exclusively to humans. Linguistically, it can encompass both humans and animals. Even if it is assumed that the term refers to humans, the applicability of Article 4(A6) is limited to participants in a levée en masse, and it does not provide a general definition of combatant status. Therefore, non-human actors would not be entitled to combatant status under this provision.

The provisions that define combatant status and membership in the armed forces do not explicitly refer to humans. The terms “person” and “inhabitants” in Article 4(A6) are not conclusive in determining whether combatant status is attributable solely to humans. The paper argues that neither the term “person” in the chapeau of Article 4 nor the term “inhabitants” can be definitively interpreted as referring to humans.

Articles 43 of AP I and Article 4 of the Geneva Conventions outline the conditions under which combatant status is granted to members of armed forces. Article 43 of AP I specifies the requirements for forming an armed group, while Article 4 of the Geneva Convention III addresses the circumstances under which irregular armed forces may be considered part of an armed force party to an armed conflict and entitled to combatant and POW status. Article 43(2) outlines exceptions to combatant status, despite membership in the armed forces, and introduces rights derived from combatant designation.

Despite the existence of various provisions addressing combatants, both Article 43 of AP I and Article 4 of the Geneva Convention III refer to “members of the armed forces” without providing a clear definition of who qualifies as a member or specifying whether this designation is limited to human beings. As a result, the question of whether combatant status is exclusively reserved for humans remains ambiguous within these legal frameworks, suggesting the potential for a broader interpretation that could include non-human entities, such as AI, in the category of combatants.

6. Personhood and Precedent: Lessons from Roe v. Wade for AI in International Humanitarian Law

The treatment of “unborn children” in landmark cases such as Roe v. Wade (1973) provides a significant precedent for the fluidity of personhood status, particularly within the context of IHL. In Roe, the U.S. Supreme Court held that unborn children could not be considered ‘persons in the whole sense’ [at 162] for the purposes of abortion. However, the Court acknowledged that unborn children could be recognised as persons for purposes outside the scope of abortion, such as in matters related to tort law or inheritance (Parness and Pritchard 1982). This complex stance from the Court signals that personhood is not necessarily a monolithic status, but can vary across different legal contexts (Hermansson 2023). This distinction supports the utility of Bundle Theory when considering the legal status of entities such as AI.

Bundle Theory proposes that personhood should not be regarded as a singular, unified concept but, instead, as a bundle of rights, duties, and recognitions that can be applied selectively depending on the legal context (Kurki 2019). This allows for the notion of legal personhood to be extended to non-human entities, such as AI, without the need for granting them full, conventional legal status. Just as unborn children are not granted full personhood in the context of abortion but may be recognised as having rights in other areas, such as property inheritance, Bundle Theory suggests that AI can be treated as a legal non-person (or person) in certain circumstances without requiring it to hold comprehensive personhood across all areas of the law.

The Roe v. Wade decision itself provides an important distinction in the way personhood is considered by the Court. While the Court concluded that unborn children did not meet the definition of “person” under the Fourteenth Amendment and therefore could not claim constitutional protections, it also acknowledged that unborn children could be recognised in other legal contexts, such ‘as acquiring rights or interests by way of inheritance or other devolution of property’ [at 162]. This distinction illustrates that the legal status of personhood can be segmented and context-dependent. The recognition that unborn children can be granted some legal rights, even without full personhood, echoes the approach that Bundle Theory advocates, where personhood is a multifaceted concept that may vary according to specific legal needs.

The correct understanding of Roe was summarised by legal scholars Parness and Pritchard, who emphasised that the Court did not prohibit lawmakers from extending personhood benefits to the unborn in other areas of law, even if it excluded them from full personhood under the Fourteenth Amendment. In fact, the Court recognised that the state had an “important and legitimate interest in protecting the potentiality of human life” and suggested that this interest could be reflected in laws that protect the unborn in areas other than abortion (Parness and Pritchard 1982). This recognition supports the idea that personhood is a flexible concept, one that can be tailored to specific legal scenarios without being rigidly applied across all legal domains. The Roe decision highlights that personhood can be a spectrum rather than a binary status, where an entity can be seen as a “person” in some respects but not in others.

Bundle Theory is valuable in this context because it separates the notion of “personhood” from the ability to hold legal rights. By adopting Bundle Theory, legal scholars and jurists can examine rights without the worry that recognising the rights of non-human entities, such as AI, automatically implies granting them full personhood. This distinction is particularly relevant in the ongoing debates surrounding the potential recognition of AI rights within IHL. In the case of AI, the question need not centre around whether AI should be treated as a “person” in the traditional legal sense. Instead, the focus should be on whether AI should be entitled to certain specific legal rights or responsibilities, such as being recognised as a combatant under Article 43(2) of AP I.

Bundle Theory’s application allows for a more flexible and nuanced approach to rights, particularly when emerging technologies such as AI are involved. It enables legal systems to explore the rights of AI without automatically extending full legal personhood to them, which could be seen as an extravagant or impractical leap. For example, one can debate whether AI, while not possessing full personhood, should be afforded specific legal entitlements such as the ability to be recognised as a combatant in IHL. By decoupling the concept of rights from personhood, Bundle Theory allows for the nuanced question of whether AI should hold particular legal entitlements, such as being protected under IHL, without implying that AI is a full legal person in every respect.

Moreover, Bundle Theory offers significant advantages when dealing with complex legal issues surrounding non-human entities. As seen in debates about AI personhood, the tendency to conflate personhood with legal rights can lead to confusion and unnecessary complications. However, if one adopts the Bundle Theory framework, the debate shifts from whether AI should be treated as a “person” to whether it should have specific legal standing, rights, or responsibilities in particular areas. In this way, Bundle Theory provides clarity by allowing for the recognition of legal rights and duties without the need to assign a blanket status of personhood.

To further illustrate the flexibility of Bundle Theory, consider its application to unborn children, particularly in the area of inheritance rights. The Roe decision confirmed that while the unborn are not persons under the Fourteenth Amendment, they can hold certain legal rights in other areas, such as the right to inherit property. This provides an example of how personhood can be tailored to particular rights, with the unborn being recognised as entitled to some rights without full personhood under the law. This concept is mirrored in the potential future treatment of AI in IHL, where AI could be granted specific rights without being fully recognised as a person, or even a person at all. For instance, AI could be entitled to legal protection as a combatant under IHL without being granted all the rights and responsibilities of human combatants.

In conclusion, Roe v. Wade serves as a crucial precedent in understanding the flexibility of personhood status in law. The Court’s decision highlights the possibility of distinguishing between personhood and the ability to hold legal rights in particular contexts. Bundle Theory extends this approach by suggesting that personhood is not a necessary condition for the acquisition of specific rights or duties. By disentangling rights from personhood, Bundle Theory allows for a more nuanced discussion of the legal status of entities such as AI, particularly in IHL. In this way, the theory offers a framework for legal systems to address complex questions about the recognition of non-human entities’ legal standing without the need to treat them as full persons under the law.

7. Argument: Autonomous Military AI’s Combatant Status Under Article 43(2) of AP I

The argument of this paper can thus be concluded as follows:

1.

If an entity fulfils the functional and operational criteria for combatant status under Article 43(2) of AP I, then it can be granted combatant status, regardless of personhood.

2.

Autonomous military AI technologies fulfil the necessary functional and operational criteria for combatant status under Article 43(2).

3.

If an entity is granted legal status and rights based on its functional role and the aggregation of relevant criteria, then the entity need not be a “person” in the traditional legal sense, as per the bundle theory of rights-based non-personhood.

4.

The bundle theory of non-personhood holds that rights and legal status are granted to entities based on the aggregation of functional roles and legal criteria, irrespective of personhood.

5.

If legal personhood is not a prerequisite for granting combatant status, then autonomous military AI technologies can be recognised as combatants, as long as they fulfil the operational criteria.

6.

Examples of non-persons that hold rights include animals (such as laws protecting certain species) and enslaved individuals (who, despite being denied full personhood historically, were recognised as having certain legal rights). These examples illustrate that entities can possess rights without being granted full personhood.

7.

Roe v. Wade demonstrates that personhood can be recognised for specific legal purposes without extending full personhood across all legal contexts (e.g., unborn children as persons for limited legal purposes, not in all contexts).

8.

Therefore, autonomous military AI technologies, which are not persons in the traditional legal sense, can be granted combatant status under Article 43(2) of AP I, as they fulfil the necessary operational criteria and, under the bundle theory of non-personhood, do not require full personhood to hold such status. Additionally, the recognition of rights in non-person entities (such as animals and enslaved individuals) further supports the possibility of AI holding rights or status under IHL.

8. Conclusions

In conclusion, this paper has articulated a central argument that the Geneva Conventions, particularly Article 43(2), do not explicitly exclude AI from being classified as a combatant or a legal person. This observation is significant, as it highlights a critical gap in the existing legal framework governing armed conflict. The traditional interpretation of combatant status has been predominantly human-centric, which may not adequately address the complexities introduced by the integration of AI technologies in military operations. By recognising that the language of the Geneva Conventions does not inherently limit combatant status to human beings, we open the door to a nuanced understanding of how non-human entities could be integrated into the legal landscape of IHL.

The application of Bundle Theory emerges as a particularly valuable framework in this context. This theory posits that legal identity can be understood as an aggregation of rights and duties rather than a singular essence tied to personhood. By employing Bundle Theory, we can conceptualise AI as a legal “non-person” that nonetheless possesses attributes aligning with the operational criteria for combatant status. This perspective allows for a more flexible interpretation of IHL, accommodating the realities of modern warfare where AI systems may perform functions traditionally associated with human combatants. However, it is crucial to acknowledge the limitations of this approach. While Bundle Theory provides a pathway for recognising AI within the legal framework, it does not fully resolve the ethical and moral dilemmas posed by the deployment of autonomous systems in combat. The question of accountability, for instance, remains a significant concern, as the delegation of lethal decision-making to machines raises profound implications for the principles of distinction and proportionality in IHL.

Moreover, as AI technologies continue to evolve and proliferate within military contexts, there is an urgent need for further exploration and potential updates to IHL. The current legal framework must adapt to the realities of AI’s role in armed conflict, ensuring that the principles of humanity, accountability, and the rule of law are preserved. This adaptation may involve re-evaluating existing definitions, establishing new categories for non-human combatants, and developing robust mechanisms for oversight and accountability in the use of AI in warfare.

In light of these considerations, it is imperative for legal scholars, policymakers, and military practitioners to engage in a collaborative dialogue aimed at addressing the challenges posed by AI in armed conflict. This dialogue should not only focus on the legal implications but also encompass ethical, moral, and operational dimensions, ensuring that the integration of AI into military operations aligns with the fundamental tenets of IHL. By proactively addressing these issues, we can better navigate the complexities of modern warfare and uphold the principles that underpin the protection of human dignity in times of conflict.