The Ethics of Responsibility in a World of Military AI

Editors' note: This essay is part of the 2026 Global Ethics Day online exclusive roundtable from Ethics & International Affairs. View the whole collection here.

The Ethics of Responsibility in a World of Military AI

Across the globe, militaries are investing growing sums in new military technology, specifically artificial intelligence (AI) and autonomous systems that accelerate lethal action. This trend is accompanied by greater geopolitical insecurity, a general sense of global peril, and a greater eagerness among actors to solve conflict by show of force. All the while, civilians continue to bear the horrific burden of seemingly endless conflicts in Ukraine and the Middle East. The moral stakes could not be higher, and ethical responsibility is needed now more than ever.

Even though there has been a solid decade, if not more, of robust and thoughtful scholarship on the ethical challenges and pitfalls of war with military AI and autonomous systems,¹ the substance of what it means to act ethically with technologies that produce accelerated lethality has slipped our grasp. There are reams of scholarship on the “responsibility gap” with AI-enabled autonomous weapon systems,² yet this gap has not been bridged or resolved in a way that takes wider notions of ethics and moral responsibility seriously. Instead, attention continues to be directed toward questions of how to operationalize conceptions of responsible technology, how to achieve responsibility by design, or how to ensure that AI lifecycle requirements can be translated responsibly into policy.³ These are important initiatives within the wider military AI debate, but they largely center on International Humanitarian Law compliance and other legal requirements an AI system must meet and, in doing so, they serve to foreground a functional technicity that coheres with the technicity of law itself. And, as we know well, not all that is lawful is necessarily ethical. Rallying around an elastic moniker such as “responsible AI” therefore runs the risk of hollowing out what it means to think ethically about the effects of new technologies in war.

Responsibility as Signifier/Responsibility and Politics

The origins of the term “responsibility” are composed of the Latin terms re (again) and spondere (to pledge). Implied in the term is therefore an obligation, an expectation, or a promise to pledge something in return, that is, to answer to an address by a person or persons. It implies an inter-personal relation, one in which all those affected by an action, such as the use of force, are considered as humans, not as abstract recipients of violence or datapoints. Within this inter-personal relationship resides the moral dimension of agency and the foundation of moral responsibility. This inter-personal grounding is not static but dynamic, requiring reflexivity about oneself and one’s embeddedness in multiple social contexts.

Politically, the term “responsibility” also has performative utility. It is a signifier that often carries its own weight without having to explain itself, and supports many interpretations. It might serve as a stand-in for ethical leadership or buttress a strategic narrative that differentiates between “responsible” (usually Western) and “irresponsible” (often China and Russia) actors. The term can apply to the technology itself, to the state using it, or to the producers making it. “Responsible military AI” refers to AI that meets certain criteria—such as trustworthiness, understandability, accountability, governability, and human-centricity. Such principles are enshrined in the UK’s 2022 policy on responsible military AI, in NATO’s 2024 Responsible Use of AI in Defence, and in the early military AI ethics guidelines adopted by the United States in 2020. What remains underspecified is how such principles are to be interpreted in practice, especially as the development and use of military AI accelerates with little restraint. A responsible state actor is one who considers the use of military AI targeting tools with conscientiousness and care, although this is often a vague normative claim. A responsible producer of AI is one who takes good care to make AI-weapon systems compliant with existing procedures and design principles across the system lifecycle, even though it stands to reason that their primary incentive will be to meet business development goals. But, once the moniker “responsible” has been bestowed, holistic ethical concerns about the use of force slip into the distance, having seemingly been managed and attenuated.

We saw this play out in the very public spectacle that pitted Anthropic against the Pentagon earlier this year, when Anthropic raised objections against the use of their technologies for surveillance of U.S. citizens or lethal targeting with fully autonomous weapons. Anthropic’s stated concern was that the technology is not yet “reliable enough.”⁴ By drawing their own two red lines, Anthropic set the ethical bar incredibly low, yet it became anointed as the ethical AI company in a sea of AI defense contractors with questionable motives.

Functionalist Responsibility

Conceptual ideas about AI and responsibility drift in meaning and significance as they circulate through political discourse. Initially, the focus was on the notion of “meaningful human control,” which carried concerns about the erosion of human moral responsibility in new military human-machine configurations. The focus soon moved toward a more abstract idea of human agency, one more concerned with “control” and less encumbered by the irresolvable and complex adjective “meaningful.” “Context-appropriate human judgement and control” has since replaced meaningful human control as a foundational pillar in discussions on human (moral) agency with AI-enabled autonomous weapon systems, subsumed under the broader category “responsible AI in the military domain.” Presently, concerns about “responsibility” are giving way to questions about the “assurance and safety” of military AI systems. With each conceptual drift, the tacit ethos underlying the concepts becomes technical and operational.

What emerges over time, then, is an increasingly functionalist approach that is concerned first and foremost with processes and operations; an approach that betrays a frustration with the fundamental non-resolvability of moral problems and dilemmas. This is also an approach in which the concern with human moral responsibility slips from our grasp. Although the original concern is a moral one, gradually resources and attention are directed toward tackling a solvable technical problem (for example, “reliable AI”). The original moral challenge of dealing with accelerated violence in war persists, but it becomes sidelined. The technical problem-solving perspective successfully diverts attention away from the ever-present demand of moral responsibility for acts of violence. All this prioritizes a conception of war not as a social and political problem, but as an engineering problem, including conceptions of responsibility in the use of force.

Responsibility and Reflexivity

None of this comes as a surprise. I have written for over a decade about how technological systems shape our subjectivities. Currently we are seeing these systems overwhelmingly pushing an image of functional-computational conceptions of ourselves and the world. Digital technologies, especially those within which individuals are intricately embedded, are seldom just a tool that we employ at will. Rather, they carry a social power. They have the capacity to subtly direct our frames of reference for decision making. In so doing, they exert a powerful—and often imperceptible—influence over our modes of governance, our security practices, our justifications for violence, and our understanding of ethics as such. But, as the use of force becomes increasingly routed through digital networks and algorithmic architectures, we must remember that these technological systems affect our thoughts and behaviors in important ways, corroding long-held humanistic values, reducing our capacity to engage in properly ethical deliberation, and, above all, eroding our understanding of moral responsibility.

The virtue ethicist Alasdair MacIntyre was acutely aware that the many social roles we inhabit can give rise to conflicts between social and private or personal moral standards. For MacIntyre, moral agency is not simply about following established moral standards. It is about assessing one’s actions against these prevailing standards. It is to engage with the tension that might arise in the conflict of competing or clashing rules attributed to different roles. To not do this means to accept or to be compliant with diminished moral agency. Since MacIntyre made these observations in the 1990s, our social roles, always mediated by technology, have only become more complex, as has our understanding of the reflexivity needed to retain moral agency. Herein resides the potential for moral responsibility and, consequently, ethical deliberations, to become hollowed out. MacIntyre’s perspective is worth quoting at length:

"Many agents exhibit no awareness of responsibilities beyond those assigned to them by their roles in each particular sphere of activity, while in their practical reasoning, they admit as their premises only those considerations sanctioned in each context by the norms defining and governing those roles. Their lives express the social and cultural order they inhabit in such a way that they have become unable to recognise, let alone to transcend its limitations."⁵

Re-Humanizing Responsibility

If this social and cultural order is one in which computational and engineering logics dominate, then this will curtail the resources available with which to understand one’s moral responsibility to reflect on one’s actions. Without these wider social and cultural resources, urgent ethical questions about the permissibility of violence, about who is responsible for creating a more lethal future, and about who will have to carry the cost of more and longer wars, will fall by the wayside in debates on military AI ethics.

Making ethical decisions remains a decidedly human affair. No degree of anthropomorphizing AI—or desiring to “program morality” into a machine—is going to change that. Ethical decision-making is often a difficult, uncomfortable, and unsatisfying task because it requires that we take responsibility for actions, especially if those actions have no clear solution and result in an adverse outcome for other beings. No part of this process can or should be sacrificed to the allure of technological progress, speed, and efficiency. Ethics is context dependent, relational, and reflexive. This deeply inefficient and inconvenient demand frustrates the logic of computational systems. But a persistent zone of moral discomfort in the use of force is essential if we are to prevent technological violence from becoming our primary mode of addressing conflict. The human edifice of ethics does not simply sustain itself. It requires constant awareness of its own fragility, situatedness, and historicity—especially in times where meaning and value are losing their footing. It is about what questions we foreground. Rather than ask “what is required to make instruments of violence legally compliant?,” we should ask “is what we are doing right now likely to propel us into a more violent future?” In a world of military AI, this is our moral responsibility.

—Elke Schwarz

Elke Schwarz is Professor of Political Theory at Queen Mary University London and currently British Academy Global Innovation Fellow at the German Council for Foreign Relations (DGAP). She has published extensively on the ethics of technology and warfare, with a specific emphasis on new and emerging military technologies, including military Artificial Intelligence (AI), autonomous weapon systems, drones and robots. She is the author of Death Machines: The Ethics of Violent Technologies (Manchester University Press, 2018). Elke is Vice-Chair of the International Committee of Robot Arms Control (ICRAC) and also an Associate with the Imperial War Museum (IWM) and an RSA Fellow. She held a 2022/23 Fellowship at the Center for Apocalyptic Studies (CAPAS) and was a 2024 Leverhulme Research Fellow.

NOTES