Re-wiring Responsibility: A European Ex-Ante Governance Framework for Algorithmic Warfare

Executive Summary

Recent years have demonstrated the rapid transformation of warfare through AI-enabled targeting and privately governed digital infrastructure. Existing legal frameworks remain tied to identifiable human intent, creating a widening accountability gap that threatens both humanitarian standards and European strategic autonomy. This paper argues that the EU must address this vacuum through geo-economic leverage. By integrating the TRACE Framework and Authority by Design into procurement regimes (EDF, EDIP, PESCO), the Union can transform accountability into a non-negotiable market condition. This approach disciplines global technology providers, protects Europe’s digital backbone, and ensures that the rules of algorithmic warfare are defined by democratic governance rather than proprietary corporate design.

Introduction

The deployment of AI-enabled targeting systems is fundamentally reconfiguring the architecture of contemporary conflict. Operational authority is increasingly mediated through opaque AI infrastructures sustained by privately governed cloud systems (Ekelhof, 2018; Andersin, 2025; Malik et al., 2025). This diffusion gives rise to ‘Structural Indifference’: a condition in which accountability is no longer concentrated in a single commander but is structurally diluted across a vast network of developers, data pipelines, and algorithmic models. 

Existing frameworks of International Humanitarian Law (IHL) and International Criminal Law (ICL) are currently ill-equipped for this systemic shift. Case studies of AI targeting systems demonstrate how targeting cycles operating at machine speed reduce human review to a mere procedural formality, making meaningful IHL assessment impossible. For the EU, this erosion of liability represents a dual crisis. Beyond humanitarian risks, the convergence of military and civilian cloud stacks means that non-auditable black boxes are being integrated into the very digital backbone that sustains Europe’s sovereign economy. Addressing this challenge requires moving beyond reactive, ex-post regulation. This paper argues that the EU’s comparative advantage lies in its role as a global regulatory power, and proposes the ‘TRACE’ Framework and the principle of ‘Authority by Design’ as tools for procurement-based governance that enable the EU to transform market access into a mechanism for enforcing meaningful human control over algorithmic warfare.

Conceptual Foundations

Authority by Design

Authority by Design (AbD) reframes accountability as a property of a system’s architecture rather than solely of the operator’s individual intent. It introduces a critical doctrinal shift: while traditional command responsibility focuses on ‘Tatherrschaft’ (the soldier’s control over the act) (Roxin, 2006), AbD identifies ’Gestaltungsmacht’: the design power of corporations to determine a system’s architectural permissions. Where foreseeable civilian harm is embedded in targeting logic or training data, the deployment of that architecture satisfies the threshold of dolus eventualis (conditional intent) (Bo, 2021; Ambos, 2005). By embedding control and auditability directly into the design, AbD ensures that liability attaches at the point of creation, before the first shot is fired. Accountability thus becomes a non-derogable infrastructural requirement, comparable to flight recorders in aviation safety.

Structural Indifference

This paper defines Structural Indifference as a condition in which technical and organisational complexity systematically erodes legal and moral accountability. In AI-enabled targeting systems, operators may formally remain ‘in the loop,’ while functionally acting as validators of machine-generated outputs under severe temporal pressure. When operators only have seconds to confirm a target, meaningful assessment under IHL becomes operationally unrealistic. Responsibility is consequently diffused across software architectures, data infrastructures, and distributed operational chains, creating an accountability erosion that existing legal frameworks struggle to address (Dickinson, 2019; Conn and Bode, 2025).

The Geo-Economic Leverage: Procurement as a Sovereign Tool

The EU’s influence over algorithmic warfare stems from its established role as a regulatory authority. By transforming accountability from a moral aspiration into a market condition, the EU can exercise regulatory leverage through procurement over global technology providers (Goussac and Boulanin, 2026).

The Dual-Use Infrastructure Trap

Global providers (e.g., Palantir, Microsoft, Google, Amazon) do not directly sell weapons, but they provide integrated dual-use infrastructures. Military AI is inextricably linked to the same cloud stacks and data pipelines used by private individuals and powering Europe’s financial systems and energy grids. This dual-use reality renders the EU AI Act’s military exemption (Art. 2(3)) a systemic security flaw (Grenda, 2025; Besliu, 2026). By leaving the military black box unregulated, the EU allows a ‘Trojan Horse’ risk to persist within its digital backbone. If architecture is deemed uncontrollable under TRACE, it becomes a toxic asset, jeopardising the provider’s standing across the EU Market. AbD ensures responsibility is defined by sovereign law, not proprietary design.

Strategic Autonomy through Standard Setting

By mandating AbD as a prerequisite for market access, the Union simultaneously secures two critical objectives. First, it prevents strategic dependency on opaque, third-party systems beyond sovereign oversight, insulating European forces from technological lock-in. Second, it triggers a ‘Brussels Effect’ in defence (Bradford, 2020): to retain EU Market access, global providers must internalise these accountability standards globally. IHL compliance thereby becomes the vehicle for European digital sovereignty, positioning the EU as the entity defining the rules of technical legitimacy for the algorithmic age.

Legal Gap Analysis: The Architecture of Impunity

The inadequacy of current legal frameworks is a direct consequence of the operational logic of algorithmic warfare. While IHL and ICL rely on human judgment and control, contemporary military AI systems are increasingly designed to bypass this human bottleneck. Three systemic failures exemplify this growing accountability gap:

• The deployment of autonomous loitering munitions, such as the Kargu-2 in Libya in 2020, undermines the foundation of command responsibility. Without auditable human authorisation for each strike, the ex post attribution of intent (mens rea) becomes impossible (UN Security Council, 2021).

• Systems such as Lavender compress targeting cycles to machine speed. When human review is reduced to only 20 seconds, legal assessments of distinction and proportionality become procedural formalities, rendering meaningful human control operationally unrealistic (Abraham, 2024; Gusterson, 2024; Mako, 2026).

• The 2026 Minab school attack in Iran illustrated the consequences of algorithmic warfare relying on outdated, incomplete, or misinterpreted data. Absent mandatory real-time verification mechanisms, AI targeting errors easily result in civilian fatalities (Kumar, 2026).

These dynamics reveal a structural mismatch: while legal doctrines remain reactive and person-centric, algorithmic warfare operates through distributed and opaque architectures that existing legal frameworks struggle to govern.

The Ex-Ante Framework: Operationalising Authority by Design

To translate abstract IHL and ICL principles into operational safeguards, this framework treats Meaningful Human Control (MHC) not as a binary condition, but as a composite function of technical, operational, and organisational constraints (Santoni de Sio and van den Hoven, 2018).  To prevent MHC from becoming a procedural fiction, this paper proposes the TRACE Framework:

MHC=f(T×R×A×C×E)

The TRACE dimensions are multiplicative: failure of a single pillar invalidates the legality of the entire system. To qualify for EU market access and defence funding, military AI systems must architecturally enforce the following requirements:

T – Temporal Validity

A strike may only be authorised once the operator has completed a verified deliberation sequence sufficient to ensure compliance with the IHL principles of distinction, proportionality, and precaution (Wolf, 2025).

• Systems must implement context-adaptive deliberation timers (e.g., 60–180s for immobile/mobile targets), ensuring human control over the attack.

• Target authorisations must be session-based, demanding renewed human validation as intelligence environments evolve.

R – Reflective Deliberation

To counter automation bias, interface design must prioritise active cognitive engagement over passive validation.

• Interfaces must require operators to classify target evidence rather than simply approve it. This prevents ‘rubber-stamping’ and places the burden of proof on the operator (Malik et al., 2025; Downey, 2025).

A – Auditability & Forensic Readiness

Accountability is contingent upon the capacity for post-event causal reconstruction.

• Every operational stage from sensor detection to kinetic release must be cryptographically signed and logged within an immutable audit log.

C – Command Override

Meaningful control requires the continuous capacity for human intervention.

• Human intervention authority must remain architecturally superior to autonomous cycles.

• If communication links are severed, systems must automatically transition into ‘Safe Deactivation’ or ‘Return-to-Base’ modes.

E – Environmental Constraints

Control must be reinforced through non-negotiable normative boundaries embedded in the system architecture.

• Mandatory ‘No-Strike’ zones (e.g., hospitals, schools, and densely populated areas) must be architecturally hardcoded into targeting logic and verified through GNSS-independent datasets, ensuring autonomous engagement is technically impossible.

• Procurement is strictly excluded for systems designed for the autonomous identification and targeting of individual human beings.

Enforcement: Consequences of Non-Compliance

Non-compliance with TRACE triggers a graduated enforcement regime. Under Regulation (EU) 2021/697, violations result in funding suspension, disbursement recovery, and debarment from EDF, EDIP, and PESCO programmes. Systemic failures attract broader penalties: providers operating dual-use architectures face fines of up to 3% of global annual turnover under the AI Act (Regulation (EU) 2024/1689, Art. 99), with designated gatekeepers facing up to 10% under the Digital Markets Act (Regulation (EU) 2022/1925, Art. 30). TRACE thus becomes a materially binding constraint across the EU Market.

Conclusion 

The future of military AI governance will not be decided in courtrooms alone, but through the governance of the infrastructures that mediate the use of force. Europe cannot prevent the emergence of algorithmic warfare, nor should it accept strategic irrelevance in an era of accelerating geopolitical competition. Its comparative advantage lies in transforming accountability into a prerequisite for technological legitimacy. By implementing Authority by Design, the EU establishes a sovereign regulatory space that disciplines global providers. This prevents opaque dual-use architectures from undermining civilian regulatory standards. Authority by Design, therefore, represents more than a compliance framework; it is a model of geo-economic governance. It ensures that European strategic autonomy remains inseparable from infrastructural accountability, and that technical responsibility is defined by sovereign law, not private boardrooms. By embedding TRACE into the 2027 EDF Work Programme, the EU can shape the global rules of algorithmic warfare before technological path dependency entrenches them.

Bibliography

Abraham, Y. (2024) ‘Lavender: The AI machine directing Israel’s bombing spree in Gaza’, +972 Magazine, 3 April. Available at: https://www.972mag.com/lavender-ai-israeli-army-gaza/ [accessed May 2026].

Ambos, K. (2005) ‘Joint Criminal Enterprise and Command Responsibility’, Journal of International Criminal Justice, 5(1), pp. 159–183.

Andersin, E. (2025) ‘The Use of the Lavender in Gaza and the Law of Targeting: AI-Decision Support Systems and Facial Recognition Technology’, Journal of International Humanitarian Legal Studies, 16(2).

Besliu, R. (2026) ‘Europe’s AI Act Leaves a Gap for Military AI Entering Civilian Life’, Tech Policy Press, 10 March. Available at: https://www.techpolicy.press/europes-ai-act-leaves-a-gap-for-military-ai-entering-civilian-life/ [accessed May 2026].

Bo, M. (2021) ‘Autonomous Weapons and the Responsibility Gap in light of the Mens Rea of the War Crime of Attacking Civilians in the ICC Statute’, Journal of International Criminal Justice, 19(2).

Bradford, A. (2020) The Brussels Effect: How the European Union Rules the World. New York: Oxford University Press.

Conn, A. and Bode, I. (2025) ‘Establishing Human Responsibility and Accountability at Early Stages of the Lifecycle for AI-Based Defence Systems’, Ethics and Information Technology, 27(4).

Dickinson, L.A. (2019) ‘Lethal Autonomous Weapons Systems: The Overlooked Importance of Administrative Accountability’, in Alcala, R.T.P. and Jensen, E.T. (eds) The Impact of Emerging Technologies on the Law of Armed Conflict. New York: Oxford University Press.

Downey, A. (2025) ‘The Alibi of AI: Algorithmic Models of Automated Killing’, Digital War, 6(9).

Ekelhof, M.A. (2018) ‘Lifting the Fog of Targeting: Autonomous Weapons and Human Control through the Lens of Military Targeting’, Naval War College Review, 71(3).

European Union (2024) Regulation (EU, Euratom) 2024/2509 of the European Parliament and of the Council on the financial rules applicable to the general budget of the Union (recast). Official Journal of the European Union, L, 23 September 2024.

European Union (2021) Regulation (EU) 2021/697 of the European Parliament and of the Council Establishing the European Defence Fund. Official Journal of the European Union, L 170, 12 May 2021.

European Union (2022) Regulation (EU) 2022/1925 of the European Parliament and of the Council on Contestable and Fair Markets in the Digital Sector [Digital Markets Act]. Official Journal of the European Union, L 265, 12 October 2022.

European Union (2024) Regulation (EU) 2024/1689 of the European Parliament and of the Council Laying Down Harmonised Rules on Artificial Intelligence [AI Act]. Official Journal of the European Union, L, 12 July 2024.

Goussac, N. and Boulanin, V. (2026) Responsible Procurement of Military Artificial Intelligence. Stockholm: SIPRI.

Grenda, K. (2025) ‘Exemption of the AI Act for military, defense and national security purposes in the light of EU law and proposed Polish regulations’, Cybersecurity & Cybercrime, 1(7).

Gusterson, H. (2024) ‘It’s All Lavender in Gaza’, Anthropology Today, 40(6).

Kumar, A. (2026) ‘Was the Attack on an Iranian Primary School a War Crime?’, Human Rights Watch, 20 April 2026. Available at: https://www.hrw.org/news/2026/04/20/was-the-attack-on-an-iranian-primary-school-a-war-crime [accessed May 2026].

Mako, G. (2026) ‘Legal Accountability for AI-Driven Autonomous Weapons’, Lieber Institute West Point, 9 March 2026. Available at: https://lieber.westpoint.edu/legal-accountability-ai-driven-autonomous-weapons/ [accessed May 2026].

Malik, A., Gul, R. et al. (2025) ‘From Human-in-the-Loop to Accountability Gaps: AI-Enabled Targeting Tools and International Humanitarian Law’, Physical Education, Health and Social Sciences, 3(4).

Roxin, C. (2006) Täterschaft und Tatherrschaft, 8th edn. Berlin: De Gruyter.

Santoni de Sio, F. and van den Hoven, J. (2018). ‘Meaningful Human Control over

Computing Systems: A Philosophical Framework’, Frontiers in ICT 5.

UN Security Council (2021) Final Report of the Panel of Experts on Libya, S/2021/229, 8 March 2021.

Wolf, R. (2025) ‘Applying Precautions in Target Verification with AI Decision Support Systems’, Israel Law Review, 58(2-3).