For much of the last decade, international lawyers have approached artificial intelligence (AI) primarily, if at all, as an object of legal regulation. The central questions have become familiar: How should international law regulate AI? Which existing legal rules apply to emerging technologies? Are new treaties required? Can established doctrines of state responsibility, international humanitarian law, human rights law, or international criminal law accommodate increasingly autonomous systems? International organisations have responded with an expanding array of broad principles, governance frameworks, and regulatory initiatives intended to ensure that AI remains compatible with established legal norms.

The terms of the debate have also shifted in another direction. Recent warnings from within the AI industry have focused on the possibility that increasingly capable systems could pose catastrophic, even existential, risks to humanity within the next decade. These warnings have intensified a familiar question: how can societies retain control over systems whose capabilities may eventually exceed the institutions designed to govern them?

These debates, despite their differences, share a common assumption: artificial intelligence is external to international law, a technological development that legal institutions must govern. This framing misses a more consequential transformation already underway. AI is becoming part of the computational infrastructure through which international legal authority is produced and exercised. It increasingly shapes what states treat as legally relevant, what evidence they consider authoritative, how risks are classified, and which forms of governmental action become possible. The question, therefore, is not simply how international law should regulate AI, but how the incorporation of AI into the exercise of sovereign power is beginning to reorganise the conditions under which international law operates.

Is AI Coming for International Law?

There is a very real risk of narrowing the debate to what AI might eventually become, rather than what computational systems are already doing within existing structures of public authority. AI does not need to become autonomous, superintelligent, or capable of eliminating humanity to transform international law. It is enough that across a growing range of domains, computational systems help identify legally relevant information, classify risks, prioritise threats, generate evidentiary assessments, and structure decision-making before formal legal judgment is made. This extends to the use of military force, which has rightly attracted considerable attention, but the same computational processes are already structuring ordinary forms of public decision-making. This transformation is particularly visible in areas traditionally associated with core functions of government. States increasingly rely on AI in sanctions administration, export controls, migration governance, investment screening, intelligence assessment, and other areas of international legal relations. Across these domains, computational systems organise the information upon which governmental and legal decisions depend. 

In these settings, algorithms rarely replace legal decision-makers. Their significance lies elsewhere. Rather, they shape what becomes legally relevant, what counts as evidence, how risks are categorised, and which forms of public authority become possible. The legal decision remains formally human; the conditions under which that decision is reached are increasingly computational. Seen in this light, the conventional framing of AI as an object of international legal regulation captures only part of the picture. A far more fundamental transformation concerns the changing infrastructure through which international law itself operates. The question, then, is what happens when not just the informational but the decisional power of states is, if not controlled, at least significantly piloted by AI: is AI an aid to decision-making or does it effectively become the decision-maker? And if that is so, how much more of a regulatory advantage does AI have over the sort of conventional regulation incarnated by international law (and perhaps all law): a regulation that does not say its name but has already seeped into the minutest crevasses of policy.

In that respect, there is a certain chutzpah in international lawyers setting out to regulate AI, when the conditions of the production of public authority have themselves already been outsourced to AI, possibly unbeknownst to many policy makers.

Fragmentation Was Never Merely Doctrinal

International lawyers often describe fragmentation as a proliferation of specialised legal regimes or competing institutional mandates. Yet the history of international law suggests a broader pattern. Fragmentation has repeatedly reflected competing political and spatial orders that organised sovereignty, legality, and universality in fundamentally different ways.

The coexistence of Western and socialist conceptions of international law during the Cold War illustrates this dynamic.* The disagreement extended beyond particular legal rules. It reflected rival understandings of the sources of legal authority, the organisation of international society, and the purposes international law was expected to serve. Similar patterns characterised earlier imperial legal orders, where international law distributed recognition, jurisdiction, and legal personality unevenly across political communities. In each case, fragmentation emerged because international law was shaped by competing political orders.

Today, however, another layer of fragmentation is emerging. The relevant competition concerns not only rival visions of international order but also the computational infrastructures through which legal knowledge is generated and sovereign power is organised. Unlike earlier forms of fragmentation, which reflected competing political and ideological visions of international order, contemporary fragmentation increasingly operates through infrastructures that shape how legal relevance, evidence, and risk are identified.

AI systems are trained on varied datasets, optimised according to different objectives, and deployed within specific institutional, political, and regulatory environments. The consequences are potentially very significant. Where states and institutions rely on systems configured around different datasets, objectives, and institutional settings, they may classify the same events, evidence, and risks in divergent ways. What follows is not merely a difference in technological capacity, but a potential difference in how risks are prioritised, governmental responses are formulated and, ultimately, legal relevance is established.

International Law’s Facticity Crisis in an Age of AI

This requires us to reconceptualise international legal fragmentation. It can no longer be understood solely in terms of competing norms, institutions, or political orders. It must also be understood in relation to the conflicting infrastructures through which states produce knowledge, classify risk, and make judgments – in short, exercise the authority that goes to the heart, historically, of their sovereignty as mediated by international law. Where different states and institutions rely on distinct computational infrastructures, they will diverge significantly, at the risk of producing tensions and even conflicts. This will not only reveal different levels of access to AI but also different levels of intensity of its use and different ways of offloading the hard work of making governance decisions to quasi-autonomous technology. What can breed divergence, of course, can also precipitate convergence as shared AI dependency creates an echo chamber of mimetic behaviour.

Still, the resulting confusion will not simply be regulatory or institutional; it can produce uneven legal-geographical conditions that destabilise the way in which international authority is constituted. The uneven configurations can be understood as “algorithmic geographies of international law”: spatial configurations of legal and sovereign authority produced through divergent computational infrastructures and the informational conditions they establish. They are unlikely to neatly overlap with states or even any recognisable legal-political authority. The concept builds on the insight that international law is constituted through geographical (North South) as well as legal (sovereignty/international community) configurations. However, it shifts attention from the conceptual maps through which international lawyers represent the field to the material infrastructures through which states and institutions produce legal knowledge, classify risk, and organise public authority.

Ultimately, it is inevitable that the very operation of AI will emerge as a fundamental threat to international law but not in the traditional way that threats to international law have been identified. This is not in the trivial sense that international lawyers will increasingly rely on AI (although there are evident risks to that too); nor is it in the sense that AI as a force might somehow be hostile to the idea of international law. It is, more subtly, because the practice that international lawyers have long grown accustomed to witnessing in states will increasingly not reliably be that of states. As international lawyers gradually come to doubt what they see as the operation of international law itself, the discipline will experience a momentous crisis of facticity. International law, just like the AI systems that increasingly feed its matrix, will begin to ‘hallucinate’, unsure about what the state practice is that it is seeing.

Just as this is likely to precipitate a major crisis of domestic governance, it could leave only a crater where international law once stood. AI is already transforming wherever computational systems shape how legal judgment is formed, governmental power is exercised, and international authority is contested. This means that AI does not merely challenge international law; it has begun to reorganise it.