By Javier Surasky
Imputability as a condition of the international order
Imputability has long been one of the enabling conditions of the international order, because it makes it possible to attribute conduct, assign responsibility, and sustain expectations of behavior in a decentralized system. The expansion of artificial intelligence into strategic functions challenges those foundations by introducing opaque, distributed decision-making processes that weaken effective control.
This post examines what happens to international responsibility when AI erodes the operational conditions under which imputability has historically functioned.
To start, we would remind that while imputability has been studied extensively in international law, it has received much less attention in International Relations, where it has generally been treated as an implicit assumption. But what if that assumption begins to collapse with the arrival and expansion of AI?
Until
recently, every international action had an identifiable actor behind it,
capable of being held accountable, if only symbolically, for its consequences.
But when AI is incorporated into strategic functions such as risk assessment or
border control, new forms of decision-making emerge that strain the traditional
framework of attribution.
As a result, a dual debate has taken shape. On the one hand, normative approaches seek to address AI responsibility by adapting pre-existing categories. On the other, some assume that AI possesses its own degree of autonomy or agency, pushing the discussion into the ethical and ontological realm and raising the question of whether the current system, “based on a paradigm of human action, is sufficient to attribute responsibility” (Human Rights Council Advisory Committee, A/HRC/60/63: para. 31).
Beyond law: imputability as systemic capacity
We choose
to distance ourselves from both approaches and think from a more uncomfortable
position, starting from the premise that the incorporation of AI systems into
strategic functions produces a qualitative transformation of imputability and
dilutes it through distributed “sociotechnical assemblages” (Domínguez
Hernández et al., 2025:43), weakening the international capacity to exercise
control and enforce accountability.
These
assemblages, where “human beings and technology interact as part of complex
sociotechnical entanglements, in which the distinction between human conduct
and machine conduct is not clear-cut” (Boutin, 2023:141), generate a new
ecology composed of a complex matrix of actors, algorithms, and machines,
making it necessary to conceptualize imputability as a systemic capacity rather
than a legal attribute.
AI and the transformation of the decision-making environment
The
starting point is a new fact: algorithmic systems are not limited to executing
predefined instructions, but participate in the production of decisions—and at
times also of actions (agentic AI)—and therefore cannot be approached as
discrete acts attributable to a specific moment. Instead, they manifest as
interactions among data, models, and infrastructures, including “nonlinear
dynamics, threshold effects, cascades, and limited predictability” (Ilcic et
al., 2025:5).
Debates on
global AI governance have addressed this shift with proposals aimed at
strengthening imputability through principles such as human-in-the-loop,
algorithmic transparency, or technical audits. All are necessary but
insufficient to establish links of attribution for conduct, as they attempt to
adapt existing models of imputability within international responsibility to
processes in which AI intervenes—presupposing that imputability can retain its
prior nature and that the challenge lies in finding the correct point of
attribution.
We do not
share that view.
Algorithmic
mediation reconfigures the conditions under which imputability operates,
producing a shift in the sociotechnical architecture of decisions in which
traditional models of responsibility cannot sustain imputability in a
politically meaningful way.
We are
facing a structural problem that cannot be resolved by simply intensifying
familiar dynamics quantitatively, but instead requires a qualitative
transformation of the conditions under which decisions operate.
This does
not deny the importance of international law or its contributions in the field
of international responsibility; rather, it points to the need to normatively
acknowledge the growing divergence between the formal attribution of
responsibility and the effective capacity for imputability over complex
decision-making processes involving AI.
The gap between legal attribution and effective control
In
international law, imputability has been built upon rules for attributing conduct to the
State, a
normative strategy that has shown resilience in changing contexts, allowing
attribution even when causal chains resulted from indirect representations.
This effect
relied on the assumption of a State with a degree of effective control over the
processes generating the conduct attributed to it—an assumption that is no
longer easy to sustain when AI systems are introduced into functions where
State agents have limited capacity to understand, anticipate, or modify AI
behavior. In an AI system, “If the operator of an AI system does not have
control over the outcome, if the machine operates largely autonomously such
that the actions or omissions of human operators are not causally linked to the
violation, it can be argued that there is no human conduct on the part of the
operator that could serve as a basis for attribution” of responsibility
(Boutin, 2023:140).
Legal
attribution may persist, but the political foundation that supported it has
fractured, giving rise to a functional gap between law and the material
conditions under which imputability can operate effectively.
Moreover,
within the international legal framework of responsibility, imputability
fulfills the political function of distributing reputational or strategic costs
among decision-makers, helping to shape expectations of future behavior. This
function collapses in AI-mediated contexts, where the will of the State agent
shifts toward complex technical processes distributed across time and space.
It is no
longer “political will,” but rather some criterion of algorithmic efficiency,
that underpins the adoption of a decision and its resulting actions, and that
efficiency criterion is constructed in a distributed manner among designers,
programmers, technology providers, and users of AI models.
Imputability, narrative, and legitimacy
A third
dimension is imputability in its narrative sense: in the international system,
responsibility co-produces a legitimate account of what happened that explains
events. But when AI intervenes, translating the technical functioning of a
system into political language means re-adapting it. The political-normative
narrative becomes a technological-explanatory one, and those who can control
that narrative are non-state actors, so imputability as public authority is
discursively deactivated.
In
conclusion, the three systemic dimensions of imputability (legal attribution,
political responsibility, and narrative control) remain, but AI profoundly
alters their modes of operation and their legitimacy.
It is
argued that AI does not introduce substantively different elements from those
imputability has already dealt with, but merely increases complexity. We see
this position as dangerous, as it frames a quantitative problem when it is in
fact qualitative, due to a transformation in the practical ontology of
decision-making imposed by AI by breaking with its grounding in discrete acts
attributable to identifiable subjects.
It is worth
noting that AI agents have evolved “from simple task executors to complex
systems capable of autonomous decision-making and multi-step reasoning, with
minimal human supervision” (ITU, 2025a:4).
It is the
algorithm that distributes the weights of the variables leading to a decision
and feeds back on its results, thereby modifying its internal structure of
weight assignment to the variables considered in each decision. The river flows
and no one bathes twice in the same waters—nor do algorithms.
Critical domains: security, sanctions, and cooperation
These transformations affect all areas of international life, but some more than others.
- The field of security and defense has been among the first to incorporate AI systems into strategic functions. From the standpoint of imputability, tensions are particularly acute here, especially when algorithmic mediation alters decision-making schemes around target prioritization or proportionality of armed responses co-produced by humans and algorithms. In such cases, a human decision-maker validates or executes a recommendation whose logic they cannot reconstruct or contest. As Australia et al. (2024: Annex, paragraph (b)) argue: “human responsibility for decisions on the use of weapon systems must be maintained, as accountability cannot be transferred to machines.” The solution applied so far has been to maintain an appearance of imputability on the human person at the cost of weakening it substantively—privileging form over substance so that it fits existing frameworks (UNIDIR, 2025; ICRC, 2025).
- In the case of economic sanctions, financial control, or the regulation of transnational flows, algorithmic mediation translates into technological optimization in terms of efficiency and objectivity (Bing Hu, 2024). But from an imputability perspective, this is the sector where decisions shift most heavily toward automated classification and scoring processes, and are therefore activated in the absence of active political decision-making. The political problem becomes a purely technical one of refining algorithms and eliminating bias, and responsibility dissolves into a technological limbo of black-box machine learning systems.
- In the realm of international cooperation, humanitarian aid, and development policies, AI systems are used to allocate resources, identify beneficiaries, assess needs, prioritize interventions, and measure impacts. The nature of the field exacerbates attribution challenges in contexts of structural asymmetry.
When cooperation includes the transfer of AI systems, recipients do not control the underlying technical infrastructures and become aligned with models, platforms, and digital developments produced in external contexts. However, under traditional logic, they remain imputable for actions taken in co-production with those systems. This creates the risk of imposing a “coloniality of imputability,” deepening regional inequality and economic polarization (ITU, 2025a:21).
All three cases, through different paths, show a decoupling effect between imputability, effective control, and accountability. The
consequences are a weakening of imputability as a disciplinary, cognitive, and
legitimizing tool, resulting in a world that is more difficult to govern
politically: under these new conditions, imputability becomes an ex post
technical-political solution associated with the mitigation, correction, and
adjustment of systems.
Winners, losers, and the future of the international order
These
transformations in international imputability produce structural winners and
losers.
The main
beneficiaries of eroded imputability are those who have greater control over
critical infrastructures (data, computing, platforms), algorithm design, and
technological discourse—namely, large transnational tech companies and States
with advanced AI capabilities.
Conversely,
States with lower technological capacity, organizations dependent on external
infrastructures, populations affected by algorithmic decisions, and even new
firms seeking to enter the digital sector face an asymmetric system of
imputability in which they must answer for decision-making systems they do not
fully control and without the capacity to counter the dominant narrative.
This
rethinking also unfolds within a broader context of institutional
reorganization of the international system, visible in processes such as UN80.
The
question we should be addressing is not how to attribute decisions mediated by
AI, but what to do when the very architecture of decision-making challenges the
conditions under which the governance of imputability has historically
operated. In that rethinking lies a significant part of the future of the
international order in the digital age: “The principles by which we want to
live must be embedded in the standards we develop” today (ITU, 2025b:7).
References
Australia,
Canada, Estonia, Japan, Latvia, Lithuania, Poland, Republic of Korea, the
United Kingdom, and the United States (2024). Draft articles on autonomous
weapon systems – prohibitions and other regulatory measures on the basis of
international humanitarian law (“IHL”) (CCW/GGE.1/2024/WP.10). Group of
Governmental Experts on Emerging Technologies in the Area of Lethal Autonomous
Weapons Systems.
Boutin, B.
(2023). State responsibility in relation to military applications of
artificial intelligence. Leiden Journal of International Law, 36, 133–150. https://doi.org/10.1017/S0922156522000607
Domínguez
Hernández, A., Perini, A. M., Hadjiloizou, S., Borda, A., Mahomed, S., &
Leslie, D. (2026). Towards a sociotechnical ecology of artificial
intelligence: Power, imputability, and governance in a global context. AI
and Ethics, 6, 43. https://doi.org/10.1007/s43681-025-00902-6
Hu, B.
(2024). AI-driven global sanctions enhancement. Frontiers in Management
Science, 3(3). https://doi.org/10.56397/FMS.2024.06.02
Human
Rights Council Advisory Committee. (2025). Human rights implications of new
and emerging technologies in the military domain (United Nations General
Assembly, A/HRC/60/63). https://docs.un.org/en/A/HRC/60/63
ICRC
(International Committee of the Red Cross). (2025). Autonomous weapon
systems and international humanitarian law: Selected issues (Position
Paper).
Ilcic, A., Fuentes, M., & Lawler, D. (2025). Artificial intelligence, complexity,
and systemic resilience in global governance. Frontiers in Artificial Intelligence, (8). https://doi.org/10.3389/frai.2025.1562095
ITU
(International Telecommunication Union). (2025a). The annual AI governance
report 2025: Steering the future of AI. https://s41721.pcdn.co/wp-content/uploads/2021/10/2502019_AI-Governance-Dialogue-Steering-the-Future-of-AI-2025.pdf
ITU
(International Telecommunication Union). (2025b). AI standards for global
impact: From governance to action. https://www.itu.int/dms_pub/itu-t/opb/ai4g/T-AI4G-AI4GOOD-2025-4-PDF-E.pdf
UNIDIR
(United Nations Institute for Disarmament Research). (2025). The
interpretation and application of international humanitarian law in relation to
lethal autonomous weapon systems: Background paper on the views of States,
scholars, and other experts. https://unidir.org/wp-content/uploads/2025/03/UNIDIR_The_Interpretation_and_Application_of_International_Humanitarian_Law_Lethal_Autonomous_Weapon_Systems.pdf