Accountability Gaps in Autonomous Warfare: Rethinking the Law of Armed Conflict in the Age of AI

Abstract

Background. Autonomous and AI-enabled weapon systems have moved from speculative debate into the center of contemporary security governance. International humanitarian law (IHL) unquestionably applies to their development and use, yet the accelerating integration of machine learning, sensor fusion, autonomous navigation, and automated target-recognition functions has exposed a persistent uncertainty: not whether law applies, but whether responsibility for unlawful harm can still be attributed in a sufficiently precise, fair, and operationally meaningful way (ICRC, 2021, 2026; UNIDIR, 2025). 
Purpose. This paper examines the accountability gaps that emerge when lethal force is mediated by autonomous or AI-enabled systems and argues that Europe requires a more explicit legal and institutional architecture than the current patchwork of general IHL rules, national weapons reviews, and non-binding AI principles. It asks how accountability diffuses across the lifecycle of such systems, why existing doctrines struggle to absorb that diffusion, and what a European response should look like. 
Methodology. The paper employs a doctrinal and policy-analytical methodology. It interprets treaty law, especially Additional Protocol I and the Rome Statute, in light of contemporary institutional practice, and it comparatively analyzes authoritative materials from the ICRC, the United Nations, UNIDIR, NATO, the European Parliament, the Council of Europe, the European Union, and SIPRI, alongside peer-reviewed scholarship on autonomy, targeting, command responsibility, and human control (Additional Protocol I, 1977, art. 36; Rome Statute, 1998, art. 28; NATO, 2024; SIPRI, 2023). 
Findings. The paper identifies five interlocking accountability gaps: an epistemic gap, a lifecycle-diffusion gap, a review-to-use gap, a command-responsibility gap, and a coalition-and-contractor gap. It argues that existing law remains applicable but is insufficiently specific to preserve meaningful accountability under conditions of algorithmic opacity, probabilistic targeting, software updates, and multinational interoperability. In response, the paper proposes a three-layer European accountability architecture combining substantive prohibitions and restrictions, operational assurance duties, and institutional oversight mechanisms. 
Research limitations. This study is doctrinal and policy-oriented rather than empirical. It does not rely on classified operational data, proprietary system-testing results, or privileged procurement documentation. Its conclusions therefore concern legal structure and governance design rather than the performance of any single platform. 
Originality/value. The paper contributes by moving beyond the binary debate between “law already suffices” and “all autonomy must be banned.” It shows that the central problem is the under-specification of accountability across the full sociotechnical lifecycle of military AI, and it develops a distinctly European pathway for preserving both operational effectiveness and legal legitimacy.