By Javier Surasky
Public International Law in the Face of the Blurring of Facts
Public
International Law (PIL) was built upon a fundamental premise: facts occur
somewhere and can be attributed to a subject. As a consequence of this
localization in space and time, they can generate normative consequences.
Territory, jurisdiction, sovereignty, and responsibility form part of the same
conceptual framework that links the possibility of anchoring a fact to a space,
an action to an author, and the allocation of consequences to a causal chain.
The
expansion of artificial intelligence (AI) pushes this framework to its limits,
not because it introduces a new type of agent—we have already discussed in
previous blog posts that AI lacks the capacity to decide or act in a legal
sense (see, for example, "AI in Three Classical Myths")—but because it
reconfigures the material and organizational conditions under which a fact is
produced.
Distributed
training, transnational infrastructures, algorithmic supply chains, automated
decisions, and cross-border effects are added to the decentralization of the
cloud, forcing PIL to rethink its answers to basic questions: where did the
fact occur? Who is its author? Who has jurisdiction? To whom can responsibility
be attributed?
Seeking
answers to these questions confronts us with a reality that is part of the
encounter between AI and PIL: "AI systems highlight the shortcomings of
existing legal frameworks and call for both innovation and discipline in their
development as they progressively extend into fields which have historically
been subject to human decision-making" (Çela et al., 2026:167).
We agree with Çela, but we believe that AI exposes a more profound crisis than the one identified there, one related to PIL's inability to operate juridically on the "fact" when AI intervenes. The legal problem does not lie, as some believe, in defining whether AI is an imputable subject, but rather in the delocalization of the fact it produces and the consequent blurring of that fact.
The Fact Itself: Localization, Attribution, and the Assignment of Meaning
PIL has not
paid sufficient attention to the fact except in relation to its legal
consequences, although it is possible to distinguish it from the act, conduct,
and factual circumstances.
The
fact—which is the focus of this blog post—is not a simple empirical event, but
rather a "legal construct," insofar as it must produce consequences that allow
PIL to consider it "legally relevant"; that is, the international legal order
itself must have granted it the capacity to produce legal effects, such as the
creation, modification, or extinction of rights and obligations.
Alongside
it appears the "act," which differs from the fact insofar as it is a
manifestation of will intended to produce legal effects. At the same time, conduct refers
to the practice of subjects with the capacity to create legal norms. Both may
take the form of actions or omissions.
Finally,
factual circumstances are those elements of reality that surround the fact, the
act, or the conduct and contribute to determining the "meaning and scope of a
given provision and, generally, its applicability to specific factual elements"
(Casanovas & La Rosa, 2018:328).
The
internationally relevant fact was historically articulated around three axes:
localization, attribution, and normative qualification. Even in complex
contexts such as cross-border operations, environmental harm, activities on the
high seas, or in outer space, PIL maintained these reference points through
legal fictions, presumptions, or special regimes. Territory, for example, may
be functionalized; attribution may be indirect; and causality may be
flexibilized—but none of this invalidates the requirement that the fact must
still occur in a place.
The legal
fact was never interpreted as a punctual event isolated from its context, but
rather as legally relevant conduct that may manifest itself as an action or
omission and take simple, continuous, or composite forms. As Shaw (2017) notes,
the international law of responsibility has operated in the face of complex
conduct, provided that such conduct could be legally delimited and attributed.
In similar terms, Sánchez Legido et al. (2022) emphasize that the
identification of the fact constitutes the logical prerequisite for attribution
and jurisdiction, even when that fact cannot be reduced to a single, isolated
act.
However, contemporary technological systems destabilize the assumption that facts can be localized in discrete actors or bounded spaces, replacing them with distributed and relational forms of agency (Arvidsson & Jones, 2023). AI thus introduces a qualitatively different difficulty, because it does not displace the fact but fragments it: training may take place in one State, deployment in another, infrastructure may belong to transnational private actors, and effects may manifest simultaneously across multiple jurisdictions. As a result, the "fact" ceases to be produced by a single agent at a given moment and instead becomes a distributed chain of operations.
International Law in the Face of Algorithmic Delocalization
The
contemporary crisis of jurisdiction is a direct consequence of the way PIL
treats the fact. The classical criteria of territoriality, nationality,
protection, and even universality, when applicable, presuppose localization;
when such localization becomes impossible or arbitrary, the jurisdictional
anchoring of PIL breaks down.
This is
precisely what occurs with complex AI systems, where the territory does not
coincide with the place of deployment, the location of infrastructure does not
coincide with the location of effects, and control does not always coincide
with benefit. The result is not a legal vacuum, but rather a potential overlap
of combined jurisdictions, with gray zones of non-control: "Global
computer-based communications cut across territorial borders, creating a new
realm of human activity and undermining the feasibility—and legitimacy—of
applying laws based on geographic boundaries" (Tzimas, 2021:230), within a
framework of asymmetries between technology, governance, institutions, and law
that "create gaps of unregulated areas which can be proven critical when we talk
about technologies that in very limited time can exponentially accelerate and
cause unexpected disruptions" (Tzimas, 2021:106).
Yet the
problem is not exclusively imputative, because it also involves a dimension
associated with the difficulty of legally localizing the fact itself. In other
words, the problem is not only assigning the fact to an entity; the problem is
isolating the fact as such. It is a question of the fact's spatiality.
We are not facing an incapacity of PIL to operate through legal fictions—an area in which it has considerable expertise. What is new is that, when attempting to identify a fact in AI systems—a prerequisite for the attribution of potential international legal responsibility—we find that its individuality has been broken, because the fact becomes the result of a distributed and indivisible algorithmic process, pushing the operational capacity of PIL into a dead end.
The "Non-Place" of the Fact: From Anthropology to Public International Law
The
difficulty of localizing the fact produced by AI systems brings their
international regulation closer to the concept of the non-place developed by
Marc Augé in anthropology. Augé notes that the anthropological place is defined
by identity, relations, and history, whereas the non-place is characterized by
their absence. In his own words, "if a place can be defined as relational,
historical, and concerned with identity, then a space which cannot be defined
as relational, or historical, or concerned with identity will be a non-place"
(Augé, 1993:82).
This
understanding of the non-place becomes more precise—and more useful for moving
from spatial place to conceptual place—if attention is paid to how space itself
is produced. As Lefebvre tells us, space is not a neutral receptacle in which
social actions unfold, but rather "a social product" and, as such, the result
of historically determined practices, relations, and forms of organization
(Lefebvre, 1974:86).
To this, we
can add what Smuha (2025:6) points out: "beyond procedures and outcomes –
sufficient attention must be paid to the social processes, structures, and
relationships that inform and are co-shaped by the functioning of such
systems."
In other
words, space does not merely "contain" action; it actively intervenes in its
production insofar as it shapes particular relations of power, encounter, and
disencounter. From this perspective, the delocalization of AI systems does not
imply the disappearance of space, but rather the production of a specific type
of spatiality that cannot be understood in classical territorial terms.
The AI
"non-place" is not, physically, the cloud, the data center, the infrastructure,
the system as software, nor its distributed action, but rather the process
through which all of these elements combine to produce a decision with
consequences in the real world. The algorithmic decision does not occur at a
single point, nor at a single moment, and therefore cannot be isolated: it
emerges over the course of an indivisible, distributed process and, in many
cases, remains hidden behind a "black box" of algorithmic reasoning. The
process as a whole lacks its own legal identity and could hardly have one,
although this is not impossible if distributed systems are met with legal
responses based on equally distributed schemes of responsibility, determined by
the production of facts, assuming that these facts are produced in a non-place
within time and space as currently captured by the legal system.
This legal
non-place does not describe the absence of facts—easily verifiable ones—but
rather the specific manner of their production in time and space, which is
incompatible with existing legal categories. PIL is thus pushed to create a
spatiality that it has not yet integrated into its corpus.
In creating
an understanding of the fact generated by AI systems, we must begin by
returning to Lefebvre (1974) and by assuming and making transparent, in
international legal terms, that the localization of the fact ceases to be a
given prior datum and instead becomes a condition produced by law itself.
As Coicaud
(2002:32–33) reminds us, politics has historically been "an enterprise of
definition and delimitation of rights, duties, and responsibilities by
territorialization (the nation–State); however, it now has to deal to a greater
extent with the processes of deterritorialization and immaterialization brought
about by the new technologies. A significant concern is how to regulate these
practices through law (…) and satisfy the demands of legitimacy. Artificial
intelligence makes this concern all the more pressing." This recalls that when
PIL fails to identify facts, attribute conduct, or exercise jurisdiction
consistently, it loses authority and legitimacy.
The proposal to move toward models of objective and collective responsibility, based on risk allocation (Finocchiaro, 2025), can be read as an attempt to restore law's operability amid the tensions posed by AI.
Open Conclusion
What AI
poses to PIL is not a technical-normative problem, but something much deeper,
insofar as it exposes a transformation at its conceptual roots: the legally
relevant fact is no longer produced in an identifiable place, but in a
processual, distributed, and transnational non-place. PIL may continue to
operate, as it always has, through fictions, presumptions, and functional
displacements—but at the cost of losing operability and legitimacy. The key
question is how PIL will confront this new reality of the non-place of fact
production without sacrificing effectiveness on the altar of power.
In these
terms, PIL faces a structural crisis resulting from the transformations
introduced by artificial intelligence, a crisis that reaches its very
foundations. If these foundations are not revised and reconfigured, the
international legal-normative order that has grown from them is destined to
progressively lose its capacity to sustain and organize the new forms of global
power.
References
Arvidsson, A. y Jones, B. (2023). International law and posthuman theory. Routledge. https://doi.org/10.4324/9781003257413
Augé, M. (1993). Los no lugares. Espacios del anonimato.
Gedisa.
Çela, E., Rao Vajjhala, N. y Aslani, B. (2026). Artificial intelligence in legal
systems: Bridging law and technology through AI. Springer.
Coicaud,
J-M. (2002). The legitimacy of international organizations. United
Nations University Press.
Finocchiaro, G. (2025). El nuevo derecho de la
inteligencia artificial. Editorial Tirant lo Blanch.
Lefebvre, H. (1974). La production de l'espace.
Anthropos.
Sánchez Legido, A.T.; Fernández Tomás, A.; Ortega Terol, J.M.;
Forcada Barona, I.; Martínez Carmena, M. y Ballesteros Moya, V. (2022). Curso
de Derecho Internacional Público (2.ª ed.). Tirant lo Blanch.
Shaw, M.N. (2017). International law (8th ed.).
Cambridge University Press.
Smuha, N.A.
(Ed.). (2025). The Cambridge handbook of the law, ethics and policy of
artificial intelligence. Cambridge University Press. https://doi.org/10.1017/9781108970607
Tzimas, T.
(2021). Legal and ethical challenges of artificial intelligence. Springer.
https://doi.org/10.1007/978-3-030-78584-3
This is the
original version of the article.
Spanish
version (ES) will be published soon.