Defence

The Autonomous Weapons Era: Military AI, Human Control, and the Governance Void

In October 2025, NATO's parliamentarians debated the governance of autonomous weapons. In January 2026, the Pentagon removed the requirement for human judgment in lethal force decisions. The governance void between those two moments defines the current era of military AI.

P J Laszkowicz

Two documents, three months apart, define the current era of military artificial intelligence.

On October 13, 2025, the NATO Parliamentary Assembly adopted Resolution 503, "Preparing NATO for the Future of Uncrewed Warfare," at its seventy-first annual session in Ljubljana, Slovenia.1 The resolution acknowledged that battlefield tactics were "rapidly evolving with increased deployment of uncrewed systems," drew lessons from the transformation of warfare in Ukraine, and called on NATO governments to develop governance frameworks for autonomous weapons that upheld international humanitarian law and the alliance's own principles of responsible AI use.2

On January 9, 2026, United States Defense Secretary Pete Hegseth released a memorandum titled "Artificial Intelligence Strategy for the Department of War," which articulated a vision of AI as the decisive advantage in great-power competition and directed the incorporation of standardised "any lawful use" language into all Department of Defense AI services contracts within 180 days.3 A subsequent memo removed the longstanding requirement — established in Department of Defense Directive 3000.09 — that operators of autonomous weapons systems be able to exercise "appropriate levels of human judgment over the use of force."4

The gap between these two documents — one calling for governance, the other dismantling it — is not a contradiction that can be resolved through better drafting or more careful diplomacy. It is a structural feature of the current international system, in which no binding legal framework governs the development and deployment of autonomous weapons, no institution has the authority to enforce one, and the states with the most advanced military AI programmes have the least incentive to accept constraints.

This article examines how that governance void emerged, what it means in practice, and why the space for closing it is narrowing.

The NATO framework

NATO's approach to military AI rests on six Principles of Responsible Use, endorsed by all allies: lawfulness, responsibility and accountability, explainability and traceability, reliability, governability, and bias mitigation.5 The principles were adopted alongside the alliance's original AI strategy in 2021 and reaffirmed in the revised strategy published in July 2024. They represent the most detailed articulation of AI ethics by any military alliance in the world.

The principles are aspirational rather than binding. "Lawfulness" requires adherence to national and international legal frameworks, including international humanitarian law — but does not define what lawful use of autonomous weapons looks like in practice. "Governability" requires that AI systems remain subject to human oversight, "enabling human actors to interact with, and where appropriate, override or deactivate AI systems, preventing unintended autonomous actions" — but does not specify how much human involvement is necessary, at what stage of the engagement process it must occur, or what happens when operational tempo makes meaningful human oversight impractical.6

Resolution 503, adopted at Ljubljana, recognised these limitations. The accompanying report by the Science and Technology Committee documented the growing role of uncrewed systems on the battlefield, drawn largely from Ukraine's experience, and identified governance and deployment challenges that the alliance had not yet resolved.7 The resolution called on NATO governments to develop regulatory frameworks, invest in counter-drone and counter-autonomous technologies, and ensure that the deployment of uncrewed systems remained consistent with international law.

But NATO's institutional structure limited what the resolution could achieve. NATO sets standards and coordinates capabilities; it does not regulate the weapons policies of individual member states. The United States, the alliance's dominant military power, was simultaneously the most advanced in military AI and the least constrained by NATO's principles. The Hegseth memoranda, released three months after Ljubljana, demonstrated that the gap between alliance governance and national action could widen at any moment — and that no institutional mechanism existed to prevent it.

The Hegseth doctrine

The January 2026 memoranda did not emerge in isolation. They were the culmination of a year-long reorientation of American military AI policy that began on President Trump's first day in office.

On January 20, 2025, Executive Order 14110 — which had established mandatory red-teaming requirements, safety protocols, and inter-agency collaboration for high-risk AI models — was rescinded.8 Three days later, Executive Order 14179 replaced it, mandating an AI Action Plan focused on "sustaining and enhancing the United States' global AI dominance."9 In July 2025, the White House released "Winning the Race: America's AI Action Plan," identifying over ninety federal policy actions organised around accelerating innovation and building AI infrastructure.10 In December 2025, a further executive order directed the Attorney General to challenge state-level AI legislation that the administration deemed "onerous."11

The Hegseth memoranda extended this trajectory from the civilian to the military domain. The vision was explicit: an "AI-first warfighting force" in which artificial intelligence was not a tool used by the military but the foundation of its competitive advantage.12 The "any lawful use" contract language served a dual purpose: it eliminated vendor-imposed restrictions on how AI systems could be used by the military, and it established the principle that if a use of force was legal under American law, the AI system enabling it should have no additional constraints.

The removal of the human judgment requirement from Directive 3000.09 was the most consequential element. The original 2012 directive, updated in 2023, had required that autonomous weapons systems be designed to "allow commanders and operators to exercise appropriate levels of human judgment over the use of force."13 This language had been interpreted as requiring meaningful human involvement in the decision to employ lethal force — what defence analysts called "human-in-the-loop" or at minimum "human-on-the-loop" control. The Hegseth memorandum reframed the standard: the relevant constraint was legality, not human judgment. If an autonomous engagement was lawful, the question of whether a human was involved in the decision was an operational matter, not a governance requirement.14

Human Rights Watch characterised the policy shift as "the US military's dangerous slide toward fully autonomous killing," arguing that removing the human judgment standard opened the door to weapons systems that selected and engaged targets without meaningful human control — systems that international humanitarian law, in HRW's view, could not accommodate.15

The international law question

The question of whether autonomous weapons can comply with international humanitarian law is not theoretical. It is the central legal issue of contemporary military AI governance, and after more than a decade of multilateral negotiation, it remains unresolved.

International humanitarian law imposes three principal obligations on the use of force: distinction — the requirement to distinguish between combatants and civilians; proportionality — the requirement that the expected civilian harm of an attack not be excessive in relation to the anticipated military advantage; and precaution — the obligation to take all feasible precautions to minimise civilian harm.16

Each of these obligations requires judgment. Distinction requires identifying who is a combatant and who is a civilian in environments where the line is not clear — where fighters wear no uniforms, where civilians may carry weapons for self-defence, where a building may be a school at one hour and a military command post the next. Proportionality requires weighing military advantage against civilian cost — a calculation that depends on context, tactical situation, and the value placed on military objectives relative to human life. Precaution requires choosing the means and method of attack that minimises civilian harm from among available options — a decision that demands real-time assessment of alternatives.17

The International Committee of the Red Cross has argued consistently that these obligations require human judgment — that the assessments involved in applying distinction, proportionality, and precaution are "complex" and depend on "the circumstances prevailing at the time of the decision to attack," and that autonomous weapons systems therefore require "that combatants retain a level of human control over their functioning in carrying out an attack."18 The ICRC's position, published most recently in an October 2025 position paper, is that autonomous weapons systems pose "serious concerns from humanitarian, legal and ethical perspectives" due to "difficulties in anticipating and limiting their effects."19

The opposing view — held implicitly by the Hegseth memoranda and explicitly by some military legal scholars — is that international humanitarian law constrains the use of force, not the means of decision-making. If an autonomous system can be demonstrated, through testing and validation, to make distinction and proportionality judgments with reliability equal to or exceeding that of human operators, then the absence of a human in the decision loop is not, in itself, a violation of IHL.20 The law requires that the rules be followed, not that they be followed by a human.

This debate has been conducted within the Group of Governmental Experts on Lethal Autonomous Weapons Systems under the Convention on Certain Conventional Weapons since 2014. In over a decade of negotiations, the GGE has produced guiding principles, working papers, and a rolling text that has been revised multiple times — most recently in May and August 2025.21 But it has not produced a binding instrument. Its mandate explicitly states that it works "without prejudging" the nature of any eventual outcome — meaning that the negotiations do not assume that the result will be a legally binding treaty.22

In September 2025, forty-two states delivered a joint statement to the GGE calling for negotiations toward a binding instrument on the basis of the rolling text.23 The ICRC and the UN Secretary-General had jointly called for the conclusion of such negotiations by the end of 2026. But the GGE operates by consensus, and the states with the most advanced military AI programmes — the United States, China, Russia, and Israel — have consistently opposed binding constraints on autonomous weapons.24 The Seventh Review Conference of the CCW, scheduled for November 2026, will determine whether the GGE's mandate is renewed, expanded, or abandoned. The prospects for a binding treaty remain uncertain.

What autonomy means in practice

The legal and diplomatic debates about autonomous weapons often proceed at a level of abstraction that obscures what is actually being deployed. The spectrum of autonomy in current military AI is wide, and the line between "autonomous" and "automated" is less clear in practice than it is in policy documents.

At one end of the spectrum are decision-support systems — AI tools that process sensor data, identify potential targets, and present recommendations to human operators who make the engagement decision. These systems are uncontroversial under existing international law: the human remains in the loop, and the AI functions as an assistant rather than a decision-maker.

At the other end are fully autonomous engagement systems — weapons that select and engage targets without human intervention after activation. No state publicly acknowledges deploying such systems, though the boundary is becoming blurred. Russia's V2U loitering munition, first deployed in late 2024 and increasingly used by early 2025, is reported to navigate GPS-denied environments autonomously, identify targets using visual terrain-matching, and potentially coordinate in swarms — with Ukrainian military intelligence reporting daily deployments of thirty to fifty units, training AI in real combat conditions.25

Between these extremes lies a range of systems with varying degrees of autonomy. Loitering munitions — weapons that can linger over a target area and strike when conditions are met — represent a growing category. Some require an operator to select the target and trigger the strike. Others can autonomously home on radar emissions or GPS signals. The market for loitering munitions was valued at approximately $530 million in 2024 and is projected to reach $815 million by 2029.26 The trend is toward greater autonomy in target identification, driven by advances in automatic target recognition and the tactical demands of contested electromagnetic environments where communications links may be disrupted.

Drone swarms — coordinated groups of autonomous or semi-autonomous unmanned systems — represent the next operational frontier. Chinese military doctrine describes drone swarms as "the next evolution in uncrewed intelligent warfare technology," with advantages in cost, scalability, and resilience over individual platforms.27 Multiple nations are investing in swarm capabilities that would allow dozens or hundreds of small autonomous systems to coordinate operations with minimal human oversight.

The practical reality is that military AI is advancing along a continuum, with each increment of autonomy making the previous governance framework slightly more obsolete. The Hegseth memoranda accelerated this process by removing the requirement for human judgment, but the underlying trend — toward faster decision cycles, more distributed operations, and reduced human involvement — predates any single policy decision.

The European position

The European Union's approach to military AI is defined by a conspicuous gap: the AI Act, the world's most comprehensive AI legislation, explicitly excludes military applications from its scope.

Article 2(3) of the AI Act states that the regulation does not apply to AI systems "placed on the market, put into service, or used with or without modification exclusively for military, defence or national security purposes."28 The exclusion is justified by reference to Article 4(2) of the Treaty on European Union, which reserves national security as a matter of member state competence, and by the argument that public international law — rather than EU regulation — is the appropriate framework for governing military AI.29

The exemption creates a paradox. Europe has built the most detailed regulatory framework for AI in the world — one that addresses bias, transparency, human oversight, and risk management with unprecedented specificity. But none of these requirements apply to the domain where the consequences of AI failure are most severe: the use of lethal force. A facial recognition system used for commercial purposes is subject to strict requirements under the AI Act. The same technology used to identify targets for military strikes is exempt.

The European Parliament has acknowledged this tension. A 2025 European Parliamentary Research Service briefing noted the "blind spots in AI governance" created by the military exemption and questioned whether relying on international humanitarian law alone was sufficient when "the pace of technological development outstrips the capacity of international legal frameworks to adapt."30 The European Security Think Tank observed that the exemption created a "regulatory oversight gap" that was particularly problematic for dual-use AI systems — those developed for civilian purposes but adapted for military applications — which fell within the AI Act's scope until the moment they were repurposed for defence.31

Individual European states are pursuing their own approaches. France's military AI strategy emphasises maintaining human control over the use of force while investing in autonomous decision-support systems. Germany's approach is cautious, reflecting public sensitivity to autonomous weapons. The European Defence Agency has launched AI initiatives focused on interoperability and common standards among member states. But no European institution has the mandate or the capacity to set binding rules for military AI that would apply across the continent.32

Who decides?

No binding international framework governs autonomous weapons. The EU has explicitly exempted military AI from its most comprehensive AI law. NATO has principles but not rules. The CCW's Group of Governmental Experts has negotiated for over a decade without producing a treaty. The United States has removed its own internal governance requirements. Russia and China participate in multilateral discussions while deploying increasingly autonomous systems.

The result is a governance void — not an absence of discussion, but an absence of authority. The question of when, whether, and under what conditions a machine should be permitted to make a decision to kill a human being is being answered, in practice, by the states and companies that build the systems, unconstrained by any binding international rule.

The space for answering this question differently — through negotiation, treaty, or norm — is narrowing. The Seventh Review Conference of the CCW in November 2026 may be the last realistic opportunity for multilateral action before the technology becomes too embedded in military doctrine to regulate. The forty-two states calling for a binding instrument represent a significant coalition, but they do not include the states whose participation is necessary for any regime to be effective.

The ICRC's call for "clear prohibitions and restrictions" by the end of 2026 reflects an understanding that governance frameworks must precede widespread deployment, because once autonomous systems are integrated into military operations at scale, the political and institutional costs of imposing constraints retroactively become prohibitive.33 The precedent of nuclear weapons — where arms control regimes were negotiated decades after the technology was deployed, and only after its catastrophic potential was demonstrated — is instructive. The question is whether the international community will learn from that precedent or repeat it.

Resolution 503 and the Hegseth memoranda represent two answers to the same question. One says: govern first, deploy within governance. The other says: deploy first, govern if necessary. The gap between them is where the autonomous weapons era is being defined — not in treaty negotiations or parliamentary resolutions, but in the decisions of military commanders, defence contractors, and political leaders who are building the future of warfare faster than any institution can regulate it.

Footnotes

  1. NATO Parliamentary Assembly, "Resolution 503: Preparing NATO for the Future of Uncrewed Warfare," 13 October 2025. ↩

  2. NATO Parliamentary Assembly, "Uncrewed Warfare — Report," Science and Technology Committee, 2025. ↩

  3. NPR, "Deadline Looms in AI Fight between Anthropic and the Pentagon," 26 February 2026. ↩

  4. Human Rights Watch, "US Military's Dangerous Slide toward Fully Autonomous Killing," 3 March 2026. ↩

  5. NATO, "Summary of NATO's Revised Artificial Intelligence (AI) Strategy," 10 July 2024. ↩

  6. NATO, "Summary of NATO's Autonomy Implementation Plan," 13 October 2022. ↩

  7. NATO Parliamentary Assembly, "Uncrewed Warfare — Report," 2025. ↩

  8. White House, Executive Order 14110, "Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence," 30 October 2023; rescinded 20 January 2025. ↩

  9. White House, Executive Order 14179, "Removing Barriers to American Leadership in Artificial Intelligence," 23 January 2025. ↩

  10. White House, "Winning the Race: America's AI Action Plan," July 2025. ↩

  11. White House, Executive Order on State-Level AI Legislation, December 2025. ↩

  12. Foreign Policy, "The Pentagon and Anthropic's High-Stakes Game of Chicken," 26 February 2026. ↩

  13. Department of Defense Directive 3000.09, "Autonomy in Weapon Systems," January 2023 update. ↩

  14. CSIS, "DOD Is Updating Its Decade-Old Autonomous Weapons Policy, but Confusion Remains Widespread," 2025. ↩

  15. Human Rights Watch, "US Military's Dangerous Slide toward Fully Autonomous Killing," 3 March 2026. ↩

  16. Additional Protocol I to the Geneva Conventions, Articles 48, 51, 57. ↩

  17. ICRC, "Autonomous Weapon Systems under International Humanitarian Law," 2025. ↩

  18. ICRC, "Position on Autonomous Weapon Systems," International Review of the Red Cross, 2025. ↩

  19. ICRC, "Autonomous Weapon Systems and IHL: Selected Issues," October 2025. ↩

  20. War on the Rocks, "Autonomous Weapon Systems: No Human-in-the-Loop Required, and Other Myths Dispelled," May 2025. ↩

  21. UN Office for Disarmament Affairs, "GGE on Lethal Autonomous Weapons Systems," 2025. ↩

  22. Ibid. ↩

  23. Stop Killer Robots, "September 2025 GGE Joint Statement," September 2025. ↩

  24. ASIL, "Lethal Autonomous Weapons Systems & International Law: Growing Momentum towards a New International Treaty," 2025. ↩

  25. MIT Technology Review, "The Future of Autonomous Warfare Is Unfolding in Europe," 6 January 2026. ↩

  26. MarketsandMarkets, "Loitering Munitions: The Convergence of AI, Autonomy, and Lethal Precision in Future Combat," 2025. ↩

  27. CNA Corporation, "PRC Concepts for UAV Swarms in Future Warfare," October 2025. ↩

  28. EU AI Act, Article 2(3). ↩

  29. EU AI Act, Recital 24. ↩

  30. European Parliamentary Research Service, Briefing on Autonomous Weapons, 2025. ↩

  31. European Security Think Tank, "Blind Spots in AI Governance: Military AI and the EU's Regulatory Oversight Gap," 3 October 2025. ↩

  32. European Defence Agency, "AI in Defence," 2025. ↩

  33. ICRC/UN Secretary-General Joint Call on Autonomous Weapons, 2023. ↩