ABSTRACT

1 INTRODUCTION

Verification of nuclear disarmament is widely recognised as a necessary condition for any credible pathway towards a world without nuclear weapons. The existing non-proliferation and disarmament framework, centred on the Treaty on the Non-Proliferation of Nuclear Weapons (NPT) and operationalised through the safeguards system of the International Atomic Energy Agency (IAEA) seeks to ensure that nuclear materials and activities remain exclusively peaceful.¹ At the same time, the strategic environment is being reshaped by a renewed nuclear arms race, the erosion of traditional arms-control instruments and the modernisation of both warheads and delivery systems by nearly all nuclear-armed states.²

These dynamics have sharpened the long-standing concern that verification arrangements for future disarmament commitments must be both technically robust and politically legitimate. Initiatives such as the International Partnership for Nuclear Disarmament Verification (IPNDV), UNIDIR’s work on verifying disarmament in the Treaty on the Prohibition of Nuclear Weapons (TPNW) and National Academies assessments of monitoring, verification and dismantlement capabilities illustrate an emerging ecosystem of research on innovative verification concepts, including the exploitation of big data, advanced sensors and remotely operated platforms.³

In parallel, rapid advances in artificial intelligence (AI), robotics and autonomy are transforming both military capabilities and verification technologies. On the verification side, the IAEA and research partners increasingly explore AI-enabled analysis of large heterogeneous datasets, as well as robotic and unmanned systems, including aerial drones, for tasks such as radiation mapping, container inspection and remote quasi-inspection of sensitive facilities.⁴ This trend is also reflected in the IAEA’s recent report to the General Conference, which highlights ongoing efforts to identify and assess emerging/innovative technologies to enhance the effectiveness and efficiency of the safeguards system.⁵ On the military side, autonomous weapons systems (AWS) have become a central focus of legal and ethical controversy. The International Committee of the Red Cross (ICRC) defines AWS as systems which, once activated, select and apply force to targets without further human intervention and has called for new internationally agreed limits to preserve compliance with international humanitarian law (IHL) and ethical standards.⁶ Within the framework of the Convention on Certain Conventional Weapons (CCW), the Group of Governmental Experts on lethal AWS has affirmed that IHL applies fully to all weapon systems and that human responsibility for decisions on the use of weapons must be retained, with appropriate

human control exercised across the life-cycle of such systems.⁷ Civil society and human rights organisations argue that AWS risks digital dehumanisation, structurally undermines accountability and should be subject to new legally binding prohibitions, especially where meaningful human control is absent.⁸

These debates intersect in complex ways with evolving nuclear delivery systems and the broader AI-nuclear nexus. Long-range missiles remain the most salient nuclear delivery platforms, but attention is increasingly directed towards advanced drones (unmanned aerial vehicles, UAVs) and other uncrewed systems. Arms-control instruments such as the Missile Technology Control Regime (MTCR) explicitly treat rockets and UAVs capable of delivering a 500 kg payload to a range of at least 300 km as potential carriers of weapons of mass destruction, while analytical work by the Nuclear Threat Initiative (NTI) and others classifies UAVs alongside aircraft, ballistic missiles and cruise missiles as nuclear-relevant delivery systems.⁹ Recent scholarship emphasises that emerging technologies, especially unmanned systems and increasingly autonomous platforms, could reshape both nuclear delivery and nuclear disarmament verification.¹⁰ In parallel, studies on military AI and nuclear risk highlight how autonomy, machine learning decision support and autonomous ISR can compress decision-making timelines and amplify the dangers of escalation in what some describe as the “third nuclear age”.¹¹ Recent UNIDIR work reinforces these concerns, noting that a number of States explicitly prioritise risks associated with the possible integration of AI into nuclear command, control and communications (NC3), and warning that such integration may exacerbate strategic-stability pressures by further compressing decision timeframes and increasing the probability of misperception or inadvertent escalation.¹²

Despite this rapidly evolving landscape, the corresponding legal discourse remains fragmented. Legal analyses of AI in the nuclear field primarily address safety and regulatory issues in the civilian nuclear sector or the compatibility of AI-assisted NC3 architectures with international law and strategic stability concerns.¹³ Work on verification focuses on institutional design and technical feasibility under the NPT and TPNW, with limited attention to how AI-enabled and autonomous tools, including drones used as verification platforms, affect evidentiary standards, due process and responsibility in disputes about compliance.¹⁴ Conversely, core AWS debates concerning meaningful human control, foreseeability, accountability and the allocation of responsibility rarely consider settings in which autonomous or semi-autonomous systems generate, process or interpret verificationrelevant data, or the specific problems posed when such systems may also be, or be closely related to, potential nuclear delivery platforms.¹⁵ This lacuna is particularly problematic given broader concerns that emerging technologies could simultaneously lower barriers to the acquisition of nuclear-capable delivery systems, including certain categories of drones, and complicate the detection and attribution of treaty violations.¹⁶

Against this backdrop, the present article examines autonomous weapons systems and related emerging technologies as they pertain specifically to nuclear disarmament verification, with particular attention to autonomous and remotely operated drones both as potential carriers of nuclear weapons and as components of verification architectures. It asks whether, and to what extent, existing international law, including nuclear disarmament and non-proliferation treaties, IHL, international human rights law, and the law of state responsibility, adequately regulates AWS (and drone-enabled verification activities) or whether additional normative development is required. Building on doctrinal and comparative legal analysis of treaty provisions, soft-law instruments, state practice in CCW and nuclear disarmament fora, and policy guidance on AI in the nuclear domain, the article argues that current law provides an indispensable but incomplete framework. In outline, it concludes that legally robust and legitimate verification arrangements for future nuclear disarmament will require explicit and operational standards on meaningful human control, transparency, auditability and data governance for AI-, drone- and AWS-enabled verification systems, together with clear rules on attribution and evidentiary reliability where such systems may also be associated with nuclear delivery capabilities. By articulating these standards, the article seeks to bridge currently separate strands of scholarship on AWS, drones and nuclear verification, and to inform ongoing practical discussions on how to design trustworthy, legally sound verification mechanisms for future nuclear disarmament.¹⁷

A small but significant body of case law—ranging from the International Court of Justice’s jurisprudence on nuclear testing and disarmament to European human-rights decisions on radiological risk and access to information and domestic and constitutional litigation on armed drones—further illustrates how courts already grapple with technologically mediated risks and access-to-justice guarantees, including, in particular, the fair-trial and equality-of-arms requirements under Article 6 ECHR and the effective-remedy standard under Article 13 ECHR (and, where EU law applies, Article 47 of the EU Charter of Fundamental Rights) even though these strands are rarely discussed together with nuclear verification.¹⁸

At the same time, questions about autonomous verification are not merely institutional or strategic. Verification findings can feed into diplomatic disputes, sanctions regimes, criminal prosecutions and domestic litigation, where they influence decisions that affect the rights and obligations of States and individuals. If the underlying evidence is generated or processed by autonomous systems, issues of chain of custody, explainability and error become directly relevant to classic access-to-justice guarantees such as the right to a fair trial, equality of arms and the right to an effective remedy. The present article, therefore, approaches the regulation of AWS- and drone-enabled verification not only as a matter of arms-control design, but also as part of a broader access-to-justice agenda, with particular relevance for European and Eastern European legal orders.

Research objectives and questions. This article pursues four objectives: (1) to map how existing international legal regimes (arms control, IHL, human rights, data protection and State responsibility) apply to autonomy and AI-enabled technologies in nucleardisarmament verification; (2) to identify the verification-relevant normative elements that recur across these regimes (e.g., accountability, due process, auditability and data integrity); (3) to assess the extent to which current treaty practice and safeguards frameworks can operationalize these elements for autonomous verification tools; and (4) to propose a legally grounded set of minimum requirements for the admissibility and contestability of machinegenerated verification evidence.

Accordingly, the article is guided by the following research questions:

• RQ1: Which existing international legal norms are most directly applicable to autonomous/AI-enabled verification technologies in nuclear disarmament?
• RQ2: What normative elements are necessary to ensure that autonomous verification outputs can be treated as reliable and legally usable evidence?
• RQ3: Where do current verification regimes (including safeguards practice) leave gaps regarding auditability, attribution and responsibility for machine-mediated findings?
• RQ4: What minimum legal and procedural safeguards are required to ensure accountability and effective remedies when autonomous verification outputs affect compliance determinations?

2 METHODOLOGY

This article adopts a qualitative, doctrinal and comparative legal research design. The core of the analysis consists of a systematic interpretation of international legal instruments relevant to nuclear disarmament and the regulation of emerging military technologies, including autonomous weapons systems (AWS), drones and AI-enabled verification tools. The doctrinal component follows the general rule of interpretation under the Vienna Convention on the Law of Treaties, combining textual, teleological and systemic approaches to assess how existing norms on disarmament, international humanitarian law, international human rights law and state responsibility apply to AWS- and drone-enabled verification activities. Particular attention is given to the evidentiary and accountability dimensions of verification, to connect the argument with broader guarantees of access to justice and fair trial rights in both international and domestic fora.

The corpus of primary materials was defined ex ante and is replicable. It consists of: (a) universal treaties and related instruments on nuclear disarmament and nonproliferation (such as the NPT and TPNW), conventional arms-control regimes relevant to delivery systems (including those covering missiles and unmanned aerial vehicles), and core IHL and IHRL instruments; (b) institutional documents, reports and technical guidance produced by the IAEA, United Nations disarmament bodies, the Convention on Certain Conventional Weapons (CCW) and other relevant international organisations concerning AWS, drones, AI and verification; (c) soft-law standards and expert principles proposed by international and regional organisations, professional associations and civil society (for example on meaningful human control, AI governance and responsible military AI); and (d) case law from international courts and tribunals—including the International Court of Justice’s jurisprudence on nuclear testing and disarmament and European Court of Human Rights decisions on radiological risk and access to information, as well as selected judgments of constitutional and supreme courts on technologically mediated evidence, armed-drone operations and military innovation where access-to-justice guarantees are at stake. Secondary materials include peer-reviewed scholarship and policy reports on AWS, drones as potential nuclear delivery systems, nuclear verification, and AI in the nuclear domain, identified through targeted keyword searches across major legal and interdisciplinary databases (including Scopus and Web of Science) and leading regional law journals. All searches were conducted in 2024–2025 using stable keyword combinations (e.g. “autonomous weapons AND verification”, “drones AND nuclear delivery”, “AI AND nuclear safeguards”, “meaningful human control AND evidence”), which are reported in the supplementary materials so that other researchers can replicate and update the corpus.

The comparative element of the study has two layers. First, it compares the treatment of AWS, drones and AI-enabled verification across relevant treaty regimes and institutional practices, mapping convergences and divergences in how notions such as human control, foreseeability, proportionality and due process are articulated. Secondly, it selectively examines practice and doctrinal debates in Eastern European jurisdictions where questions of technologically mediated evidence, military use of AI and drones, or nuclearrelated risks have prompted legislative or judicial responses. The analysis proceeds in three steps: 1) identification and classification of legal provisions and policy statements that bear on AWS- and drone-enabled verification; 2) reconstruction of their underlying normative assumptions, especially regarding human control, accountability and evidentiary reliability; and 3) evaluation of these assumptions against access-to-justice standards - in particular fair-trial and effective-remedy guarantees under international and regional human-rights instruments, leading to the normative proposals developed in the Results and Conclusions section. With a particular focus on the Council of Europe framework (ECHR Articles 6 and 13) and the EU standard of effective judicial protection (Charter Article 47), insofar as verification outputs may be used in proceedings or regulatory decisions within European jurisdictions. The tables and figures presented in the Results section, therefore, synthesise the material in a qualitative, interpretive manner rather than reporting a separate statistical dataset.

No separate empirical dataset involving human subjects, experimental measurements or proprietary computer code was generated for this research; instead, the study relies on doctrinal and policy analysis of publicly available legal and technical materials. All materials consist of publicly available legal and policy documents and published scholarship. Where access to materials depends on subscription-based databases or publisher paywalls, full citations are provided in the article so that readers with institutional access can consult the same sources. A consolidated list of all primary instruments and decisions, together with the keyword strings and inclusion/exclusion criteria used to construct the corpus, will be made available as supplementary material or via an open institutional repository, subject to the copyright and licensing conditions of the underlying documents. This enables other scholars to replicate the selection and to build upon the analytic framework proposed in this article.

Generative artificial intelligence (GenAI) was employed only as an auxiliary instrument in the research and writing process. Specifically, ChatGPT (OpenAI) was used to assist in: (a) preliminary mapping of existing literature and policy documents by suggesting additional search terms; (b) generating alternative formulations and structural options for certain sections; and (c) language refinement and consistency checks in the drafting stage. All legal interpretations, argumentation and conclusions were developed independently by the author, who reviewed, verified, and, where necessary, substantially rewrote any AI-assisted draft text. No AI tool was used to make legal or policy decisions, to generate original research findings, or to determine the normative positions advanced in the article.

The authors bear sole responsibility for the accuracy of citations, the selection and interpretation of sources, and all substantive claims contained in this work. The descriptive distributions and matrices in the Results section are based on this manual doctrinal and policy analysis and should be read as qualitative syntheses, not as the product of AI-driven document coding or statistical modelling.

3 RESULTS

The findings of this study can be organised into three interrelated clusters, which correspond to: 1) the extent to which existing international law regulates autonomous and AI-enabled technologies in nuclear disarmament verification; 2) the specific position of drones as both verification platforms and potential nuclear delivery systems; and 3) the emergence of a common set of normative standards that could guide the design and use of such technologies in legally robust verification architectures. Tables 1–4 and Figures 1–3 provide structured overviews of the doctrinal and policy-mapping results referred to below.

First, the cross-regime coding exercise shows that the bulk of relevant provisions in nuclear disarmament treaties, safeguards practice, international humanitarian law (IHL), human rights law and the law of state responsibility are formulated as obligations of control and due diligence, while a significantly smaller proportion directly addresses accountability and evidentiary reliability (Table 1).

*Coded categories: Control & due diligence, Accountability, Evidentiary reliability

A qualitative review of these provisions indicates that most are framed as obligations of control and due diligence, with a smaller cluster concerned with accountability mechanisms and an even more limited subset explicitly addressing evidentiary reliability. Figure 1 offers a stylised representation of this pattern.

This distribution confirms that existing regimes were not drafted with autonomous or AI-enabled verification tools in mind: they provide clear expectations that states must exercise adequate control over weapons and related technologies, but they offer little explicit guidance on how to treat data generated by autonomous systems, how to ensure its auditability, or how to allocate responsibility when such systems malfunction or embed bias. The gap is particularly evident when these general norms are read alongside contemporary debates on autonomous weapon systems (AWS), where the ICRC and others have called for new legally binding rules to address unpredictability, human control and responsibility in the use of autonomous force.¹⁹ The findings therefore indicate that, while AWS debates already recognise the need for technologically specific limits, analogous questions arising in the context of nuclear disarmament verification remain only partially articulated in existing instruments.

Second, the technical and policy mapping demonstrates that drones occupy a particularly sensitive dual-use position. On the one hand, safeguards-oriented work by laboratories and international partners shows that unmanned aerial vehicles (UAVs) can greatly enhance verification by enabling remote visual inspection, environmental sampling and radiation mapping in inaccessible or hazardous locations.²⁰ On the other hand, both technical reports and recent analyses of nuclear safety and security highlight that many of the same UAV configurations—especially long-range systems with modular payload bays—possess characteristics comparable to delivery platforms for weapons of mass destruction or can be repurposed for hostile operations against nuclear facilities.²¹

A structured summary of technical characteristics relevant to verification is provided in Table 2, while Figure 2 visualises this overlap, showing that some archetypal verification drones fall within recognised missile-technology thresholds or could be modified to do so. The findings thus confirm that drones deployed in verification roles cannot be treated solely as benign technical aids; they must also be viewed as potential objects of regulation in their own right, raising proliferation, targeting, and escalation concerns that intersect with their verification functions. Related case law on armed-drone operations—such as the German Ramstein litigation, United States federal cases in the Al-Aulaqi line and criminal proceedings involving planned drone-delivered attacks—underscores that courts already treat certain categories of drones as inherently dangerous weapons and scrutinise the adequacy of legal frameworks governing their use, including in terms of extraterritorial obligations and access to effective remedies.

Note. Circle A (left): UAVs deployed or proposed for verification/monitoring tasks (long loiter time, high-resolution sensors, secure communications). Circle B (right): platforms meeting missile-related payload/range thresholds for WMD delivery (payload ≥ 500 kg and range ≥ 300 km, or equivalent criteria). The intersection represents long-range, modular verification UAVs with potential delivery capability.

Third, the analysis of state submissions to the Convention on Certain Conventional Weapons (CCW) Group of Governmental Experts on lethal AWS, ICRC guidance, NGO advocacy and specialist safeguards literature reveals a converging set of normative elements that could serve as building blocks for regulating autonomous and AI-enabled verification systems. These elements include requirements for meaningful human control over critical functions, limitations on unpredictability, transparency and explainability of system behaviour, robust data integrity and cybersecurity safeguards, and pre-deployment technical-legal reviews. 22 The frequency distribution of these elements is displayed in Figure 3, while a consolidated matrix of recurring normative themes is presented in Table 3.

Note: Entries “high”, “medium” and “low” reflect the author’s qualitative assessment of the prominence of each element in the sampled materials rather than the output of a formal quantitative coding exercise.

Meaningful Human Control in verification. For the purposes of nuclear disarmament verification, meaningful human control (MHC) should be understood as the continuous and effective involvement of designated human officials in the critical verification functions that can produce legal consequences. Unlike the use-of-force context, where MHC primarily concerns human control over target selection and the application of force, in the verification context, MHC concerns human control over evidence generation and compliance assessment. At minimum, MHC requires: 1) human authorisation of mission objectives, geographic scope and rules of operation; 2) practical ability to supervise operations and intervene/override or abort autonomous behaviour; and 3) human validation and formal sign-off of any machine-generated compliance-relevant assessment before it is treated as evidence or triggers enforcement measures. MHC, therefore, depends on auditable logs, explainability sufficient for review, and an institutional review pathway that can correct or contest autonomous outputs.²³

At the same time, studies on AI in nuclear safeguards underline that AI-based tools are already being explored for anomaly detection, pattern recognition and the fusion of geospatial and operational data, while stressing that these developments must be aligned with evolving AI regulatory frameworks and safeguards-specific legal constraints. 24 The findings of this study systematise these strands by showing that, across the examined

discourse, there is implicit agreement on a core normative grammar-human control, transparency, auditability, data protection, but that explicit, verification-specific operationalisation of these principles is still largely absent.

Finally, the jurisprudential survey summarised in Table 4 shows that courts and quasijudicial bodies have begun to engage with technologically generated evidence and with disputes arising from nuclear-related risks and armed drone operations. One cluster of decisions, exemplified by Prosecutor v Ahmad Al Faqi Al Mahdi (ICC), Big Brother Watch and Others v the United Kingdom (ECtHR), City of Sioux City v Jacobsma and State v Loomis, addresses satellite imagery, bulk interception, automated enforcement and algorithmic risk scoring, and thus illustrates how judges assess the admissibility, probative value and contestability of machine-generated evidence.

A second cluster, including the ICJ’s Nuclear Tests, Nuclear Weapons and Marshall Islands cases, ECtHR judgments such as McGinley and Egan, LCB and Roche, and domestic and constitutional litigation on armed drones in Germany (Ramstein) and the United States (the Al-Aulaqi line of cases), concerns the evidentiary and remedial dimensions of nuclear testing, radiological exposure and lethal drone strikes.

Taken together, these strands reveal both a willingness to admit technologically mediated evidence and a persistent concern with issues such as chain of custody, access to underlying data, the burdens of proof and the effective availability of remedies.

While none of the decisions examined directly addresses autonomous nuclear disarmament verification, they illustrate the types of procedural safeguards that will be necessary if autonomous or AI-enabled verification outputs are used to support allegations of noncompliance or to justify sanctions and other coercive measures, and they show how courts link technologically mediated risk to fair-trial guarantees, equality of arms and the right to an effective remedy. Against the background of broader analyses of AI, strategic stability, and nuclear risk—which emphasise compressed decision-making timelines, new escalation pathways and uncertainties in human-machine interaction—the study concludes that integrating autonomous and AI-enabled systems into nuclear disarmament verification without explicit evidentiary and accountability standards would risk exacerbating, rather than reducing, doubts about compliance and fairness.²⁵

Taken together, the findings support three core conclusions: 1) existing law forms a necessary but incomplete basis for regulating autonomous verification technologies; 2) drones used for verification must be governed with full awareness of their dual-use potential as nuclear-relevant delivery systems; and 3) future nuclear disarmament arrangements will need to codify verification-specific standards on human control,

transparency, data governance and evidentiary treatment if they are to maintain both technical effectiveness and legal legitimacy.

As explained in the Methodology section, the corpus combines treaty provisions, institutional reports, technical specifications, expert statements and case law, and the Results section draws on this mixed set of sources for cross-regime comparison.

4 DISCUSSION

The findings of this study broadly confirm the initial hypothesis that existing international law offers a necessary but incomplete framework for governing autonomous and AI-enabled technologies in nuclear disarmament verification. At the level of principle, norms on state responsibility, due diligence, weapons reviews and IHL rules on the conduct of hostilities already impose robust expectations of human control, predictability and accountability in the use of weapons systems.³⁷ However, when these norms are applied to verification tools such as autonomous data-collection platforms, AI-driven anomaly detection, and dronebased inspection systems, significant gaps emerge regarding evidentiary standards, the auditability of algorithmic outputs, and the allocation of responsibility for technically mediated non-compliance assessments. This result complements the ICRC’s call for new legally binding rules on autonomous weapon systems and meaningful human control, while extending the analysis beyond the battlefield to the verification infrastructures that underpin nuclear disarmament regimes.³⁸

Cyber resilience, spoofing, and legal responsibility. Autonomous verification drones rely on geolocation and time-stamping to link observations to treaty-relevant coordinates. Coordinate manipulation (e.g., GNSS spoofing) can produce technically “plausible” outputs while attaching them to the wrong location and/or mission context, thereby compromising the chain-of-custody and the reliability of compliance-relevant evidence. This creates a compounded attribution problem: it becomes difficult to distinguish an actual violation from a corrupted verification record, and the technical attribution of the interference itself is also complicated. From a legal perspective, this strengthens the case for 1) explicit integrity and cyber-resilience obligations in verification arrangements (tamper-evident logging, authenticated telemetry, redundancy/corroboration, incident disclosure and forensic access) and 2) procedural safeguards before autonomous outputs can trigger legal effects, including a structured opportunity to challenge both the data and the integrity of the collection process.

From the standpoint of access to justice, these gaps create uncertainty about how parties can meaningfully contest machine-generated verification findings, demand disclosure of

underlying models and data, and obtain effective remedies when autonomous systems contribute to erroneous or discriminatory outcomes.

The case law surveyed in this article confirms that these concerns are not hypothetical. In the nuclear field, ICJ and ECtHR decisions on atmospheric testing, radiological exposure and access to information (Nuclear Tests, Nuclear Weapons, McGinley and Egan, LCB, Roche) show how evidentiary uncertainty, secrecy and limited access to technical data can shape the justiciability of claims and the availability of remedies. In parallel, constitutional and administrative litigation on armed-drone operations in Germany (Ramstein) and the United States (the Al-Aulaqi cases) illustrates both the potential and the limits of judicial review where lethal force is exercised through remote or semi-autonomous platforms and key operational information remains classified. From an access-to-justice perspective, these strands reinforce the need for verification architectures that generate auditable records, allow affected parties and courts to reconstruct critical decision points, and provide sufficiently transparent explanations of how autonomous and AI-enabled systems contributed to compliance assessments.

In comparative perspective, these findings resonate with, but also refine, earlier work on AI, autonomy and strategic stability. The SIPRI series on artificial intelligence and nuclear risk has shown that AI-enabled systems may compress decision-making timelines, introduce new escalation pathways, and complicate crisis management, especially in nuclear command, control, and communications and on advanced delivery platforms.³⁹ By contrast, the present study concentrates on the upstream moment in which compliance with disarmament or non-proliferation obligations is assessed, documented and, ultimately, contested. It suggests that many of the destabilising features identified in the AI/strategicstability literature – opacity of complex models, difficulty of validating performance across contexts, and the risk of automation bias-also arise when autonomous or AI-enabled systems are embedded in verification architectures. If verification outcomes are increasingly shaped by systems whose functioning is not transparent to inspectors, diplomats or courts, the credibility of assurances and the perceived fairness of compliance procedures may be undermined even in the absence of any material breach.

The dual-use role of drones is a central illustration of this tension. Technical work by Pacific Northwest National Laboratory has demonstrated that unmanned aerial vehicles (UAVs) can significantly enhance IAEA-type safeguards through remote imaging, environmental sampling and access to hazardous or otherwise unreachable locations.⁴⁰ At the same time, policy analyses on the future of IAEA safeguards underline the importance of preserving confidence in verification conclusions and managing perceptions of politicisation or technological overreach. Parallel arms-control discussions point out that many UAV configurations have characteristics comparable to recognised delivery systems and therefore

raise questions about their regulation in zones free of weapons of mass destruction and their delivery vehicles.⁴¹ Recent commentary on drone technology and the future of nuclear weapons further emphasises that AI-enabled drones may blur the line between strategic delivery platforms and ostensibly defensive or verification-oriented systems.⁴² Taken together, these strands reinforce the study’s conclusion that drones deployed in verification roles cannot be treated as neutral technical enablers; they must be governed as potential vectors of nuclear risk in their own right, with appropriate limits on range, payload, autonomy and data-handling functions.

The emerging normative template identified in the findings also aligns with broader civilsociety and expert discourse on autonomous weapon systems. ICRC instruments, UN disarmament debates, and NGO advocacy repeatedly call for meaningful human control, limits on unpredictability, and robust legal review of new weapons, while emphasising that accountability cannot be transferred to machines. Human Rights Watch’s recent report on autonomous weapons and digital decision-making adds a human rights perspective, arguing that delegating life-and-death decisions to opaque systems jeopardises rights to life, non-discrimination, and an effective remedy, and therefore warrants a legally binding instrument.⁴³ The present study suggests that the same normative elements-human control, transparency, predictability and access to remedy- should be systematically transposed into the design, deployment and legal assessment of autonomous or AI-enabled verification systems, including those based on drones. For courts and compliance bodies, particularly in Eastern European jurisdictions where ECHR standards (Articles 6 and 13 ECHR) structure domestic procedure, these elements translate into concrete doctrinal requirements that evidence derived from autonomous verification chains remains intelligible, contestable and accompanied by clear avenues of appeal. Doing so would help ensure that technologically sophisticated verification does not erode, but rather reinforces, the rule-oflaw foundations of disarmament processes.

In practical terms, the findings indicate that future nuclear disarmament and nonproliferation arrangements should incorporate verification-specific obligations on data governance, algorithmic transparency and evidentiary treatment. Work on AI for nuclear safeguards verification by Oak Ridge National Laboratory already highlights both the promise and the risks of introducing machine-learning tools into safeguards practice, stressing the need for careful integration with existing legal and institutional standards.⁴⁴ Building on this and on NTI-sponsored analyses of the future of IAEA safeguards, states and international organizations could, for example, require: 1) predeployment legal-technical reviews of verification algorithms and autonomous

platforms; 2) auditable logging of system behavior and data transformations throughout the verification chain; 3) clear rules on when and how machine-generated evidence may be used to support non-compliance findings; and 4) mechanisms for independent expert challenge and review of contested technical assessments.⁴⁵ For drones specifically, verification mandates might include explicit constraints on operational envelopes and payload configurations, integration of cybersecurity and data-protection standards, and measures to separate verification UAV fleets from systems designed or perceived as potential nuclear delivery platforms.

At the same time, several limitations of the present study must be acknowledged. First, the analysis is primarily doctrinal and policy-oriented: it relies on open-source legal instruments, technical reports, policy papers, and expert commentary, and thus cannot capture classified or state practice materials that may significantly shape how autonomous and AI-enabled systems are actually being integrated into nuclear infrastructures. Second, the coding of legal and policy documents, particularly the identification of core normative elements, necessarily involves a degree of interpretive judgement, even though it was guided by existing syntheses on AI, AWS and strategic stability.⁴⁶ Third, the study is technology-sensitive but does not undertake independent technical validation of specific algorithms, platforms or deployment concepts; its conclusions about risk and responsibility are therefore conditional on the current state of public technical knowledge, which is itself rapidly evolving. Finally, while the discussion incorporates global perspectives, it inevitably reflects the biases of the English-language literature and of debates taking place in UN and Euro-Atlantic forums.

These limitations point directly to future research needs. One priority is empirical, interdisciplinary work that brings together lawyers, computer scientists and verification practitioners to model concrete verification scenarios involving autonomous systems and drones, including stress-testing proposed standards for human control, transparency and auditability. Another is region-specific analysis of how autonomous verification tools might be perceived and regulated in different strategic environments, for example, within prospective WMD-free zones in the Middle East or in regions highlighted in SIPRI’s East Asian and South Asian volumes on AI and nuclear risk.⁴⁷ Region-specific perspectives are particularly important. For example, UNIDIR’s recent regional analysis of AI-driven threat perceptions in the Middle East WMD-Free Zone context illustrates how the same enabling technologies can be interpreted either as drivers of arms-racing dynamics or as potential

facilitators of risk-reduction and verification arrangements, depending on regional security conditions and negotiating priorities.⁴⁸ A third direction is to explore how evidentiary doctrines in international and domestic courts can adapt to machine-generated verification outputs without sacrificing principles of adversarial challenge, due process and public reason-giving. Finally, as both AI and drone technologies continue to develop, longitudinal studies will be needed to track whether the integration of autonomous systems into verification practices actually enhances confidence in compliance or instead generates new forms of contestation and mistrust. The present study offers an initial legal framework for this inquiry, but sustained engagement by states, international organizations and scholars will be required to translate its proposals into operational, treaty-ready norms.

5 CONCLUSIONS

This article set out to answer a straightforward but largely neglected question: how do autonomous weapons systems, drones and other AI-enabled technologies fit into the legal framework of nuclear disarmament verification? To address it, the study used doctrinal and comparative legal methods. It systematically examined treaty law on disarmament and non-proliferation, international humanitarian and human rights law, the law of state responsibility, soft-law standards on autonomous systems and AI, state practice in multilateral forums, and selected case law on technology-generated evidence. Rather than proposing entirely new concepts, the method was to read these existing sources together and to test how far they can be stretched to govern autonomous and semi-autonomous verification tools, including drones that are technically capable of acting as nuclear delivery platforms.

The main results can be summarised quite concretely. First, current international law does reach autonomous and AI-enabled verification systems, but only in a general and fragmented way. It speaks clearly about state control, due diligence and responsibility, yet remains largely silent on how to treat machine-generated verification data as legal evidence, how to ensure its auditability, and how to allocate responsibility when automated systems influence compliance assessments. Second, drones emerge as a genuinely dual-use technology: the same classes of unmanned aerial vehicles that can greatly improve verification by extending the reach of inspections also share key technical parameters with platforms that raise proliferation and deterrence concerns. Third, across ICRC guidance, CCW debates, safeguards practice and AI-governance discussions, a fairly stable cluster of normative building blocks can be identified-meaningful human control, limits on unpredictability, transparency and explainability, data-integrity and cyber-security

safeguards, and pre-deployment legal-technical review–even though these elements are rarely translated into detailed, verification-specific standards.

On this basis, the article’s central claim can be confirmed. Existing law provides a necessary foundation, but it is not sufficient on its own to guarantee that autonomous and AI-enabled verification will be both effective and legitimate. The contribution to legal scholarship lies in three clarifications. First, it shows that the AWS debate cannot remain confined to battlefield uses: many of its core concepts, especially meaningful human control and accountability, must be carried over into the design and legal evaluation of verification chains. Second, it reframes drones used for monitoring not only as instruments of verification but also as objects that require careful legal classification due to their proximity to recognised nuclear delivery systems. Third, it organises dispersed principles from disarmament law, AI governance, and human rights practice into an integrated set of criteria that can guide future regulation of autonomous verification technologies. In this way, the study slightly shifts the state of knowledge: the problem is no longer whether law applies to such systems, but rather what additional, verification-specific standards are needed for evidence, responsibility, and the management of dual-use platforms.

The practical and forward-looking implications follow directly from this. In terms of implementation, the analysis suggests that future nuclear disarmament agreements and safeguards arrangements should move beyond technology-neutral formulations and incorporate explicit clauses on: (i) minimum requirements for human control over critical verification functions; (ii) transparency and logging of algorithmic processes that shape verification conclusions; (iii) secure and traceable handling of data produced by autonomous sensors and platforms; and (iv) the special treatment of drones whose technical characteristics bring them close to delivery systems regulated in other arms-control regimes. These elements can be developed through interpretative understandings, verification protocols, institutional review procedures and model guidelines for the design and certification of autonomous verification tools.

At the same time, the work points to clear avenues for further research and gradual implementation. Interdisciplinary projects can test proposed standards in simulated verification scenarios, helping to translate broad legal principles into technical specifications and test procedures. Regional case studies can explore how different security environments-particularly those where nuclear risks and unmanned systems are already closely intertwined- might adapt and refine these standards. Courts and compliance bodies, in turn, will need to experiment with evidentiary rules that allow for the use of machine-generated verification outputs while preserving due process and meaningful avenues for challenge. The jurisprudence reviewed in this article— from ICJ and ECtHR cases on nuclear risk and radiological exposure to domestic litigation on armed drones and algorithmic decision-support—already sketches the contours of such rules, even if it has not yet been systematically connected to the design of nuclear disarmament verification regimes. Taken together, these steps outline a realistic path for

integrating autonomous systems, drones, and AI into nuclear disarmament verification that strengthens, rather than weakens, both legal certainty and trust between states. At the same time, by embedding access-to-justice safeguards—fair-trial guarantees, robust evidentiary standards and effective remedies—into the regulation of autonomous verification, these measures can help ensure that technologically sophisticated compliance mechanisms reinforce, rather than erode, the ability of individuals and States to obtain justice in disputes over nuclear disarmament obligations.

FOOTNOTES

REFERENCES

AUTHORS INFORMATION

RIGHTS AND PERMISSIONS

EDITORS

ABOUT THIS ARTICLE

DETAILS FOR PUBLICATION

FUNDING

AI DISCLOSURE STATEMENT

АНОТАЦІЯ