1. Introduction
It is almost ritualistic to begin any discussion of modern international law and legal order with the Treaty of Westphalia. Yet discussion of this cornerstone of international law often overlooks the intent of the treaty—the codification of the prohibition on intervention.
International law prohibits intervention by a State in the internal or external affairs of another State (Nicaragua v United States of America, 1986, p. 202; United Nations General Assembly, 1970). The prohibition on intervention, which forms “part and parcel of customary international law” (Nicaragua v United States of America, 1986, p. 202), is the “corollary of every State’s right to sovereignty, territorial integrity and political independence” (Jennings and Watts, Reference Jennings and Watts1992, p. 428; Nicaragua v United States of America, 1986, p. 202).
Generally speaking, a prohibited intervention is one that interferes by coercive means, either directly or indirectly, in matters which a State is permitted, by the principle of State sovereignty, to decide freely (Nicaragua v United States of America, 1986, p. 205). Such matters include, for example, a State’s economic, political, social, and cultural systems, and foreign policy (Nicaragua v United States of America, 1986, p. 205). Intervention will be “coercive” where it deprives, or is intended to deprive, another State of its ability to control, decide upon, or govern the matter in question, either by compelling the State to act, or refrain from acting, in a particular way (Schmitt, Reference Schmitt2017, p. 317; Jennings and Watts, Reference Jennings and Watts1992, pp. 430–31; United Nations General Assembly, 1970).Footnote 1 Although the prohibition on intervention is clearly established in international law, its precise scope and content, including what constitutes “coercion” or the “internal or external affairs” of a State, remains unclear. This is due, in large part, to a lack of international jurisprudence and the fact that State practice is limited, varied, and often kept secret. This poses significant challenges for determining the scope and content, as well as the practical application of, the prohibition on intervention, not least in the cyber context. This complexity is compounded by divergent State positions on the legal status of sovereignty in cyberspace. The United Kingdom and, more recently, the United States (2020 speech of the Department of Defense General Counsel) have characterised cyber “sovereignty” as a general principle of international law rather than an independently breachable rule. By contrast, France, the Netherlands, and several other States have expressly endorsed sovereignty as a rule capable of violation through cyber operations. This doctrinal divergence is not novel, nor does it displace the relevance of the prohibition on intervention. The 2015 UN Group of Governmental Experts (GGE) consensus norms and the subsequent 2021 GGE and Open-Ended Working Group (OEWG) reports explicitly affirm that the principle of non-intervention applies to Information and Communication Technologies (ICT) enabled activities. Together, these instruments provide authoritative—albeit soft-law—confirmation that coercive interference with a State’s reserved domain remains unlawful irrespective of the unresolved debate over sovereignty’s precise doctrinal status.
This article outlines the sources and elements of prohibited intervention and the distinction between prohibited intervention and other forms of pressure, interference, or unfriendly acts. The article also sets out a range of factors, which may assist in determining the extent to which a State’s conduct is either more or less likely to breach the prohibition on intervention and the application of these factors to a number of hypothetical scenarios. In particular, the article considers how artificial intelligence (AI) is reshaping the landscape of coercion. Emerging AI-enabled operations—such as deepfakes, data poisoning, and algorithmically amplified influence campaigns—challenge traditional understandings of sovereignty and non-intervention. These techniques do not rely on force, but may nonetheless erode a State’s capacity to make autonomous decisions by distorting its cognitive or informational environment. Drawing on the Tallinn Manual (Schmitt, Reference Schmitt2017, p. 316) and Milanovic’s models of coercion (Milanovic, Reference Milanovic2023, pp. 601, 609), this article argues that the legal test for prohibited intervention must evolve to include such digitally mediated harms. As such, this article offers a unique model, developed by the author, called the Decisional Integrity Model.
The article proceeds in four parts. Part one traces the evolution of the principle of non-intervention, emphasising its conceptual roots in sovereignty and its doctrinal development through the Nicaragua judgment and subsequent legal instruments. Part two explores how AI and digitally mediated operations challenge the foundational assumptions of that principle—particularly by shifting coercion from physical to cognitive domains. Part three introduces the Decisional Integrity Model, a five-part legal test for identifying when AI-enabled influence operations rise to the level of prohibited intervention. Finally, the conclusion reflects on how international law must evolve to safeguard not only territorial sovereignty, but the epistemic conditions under which sovereign choices are made.
2. The interplay of sovereignty and intervention
Sovereignty is the conceptual cornerstone of modern international law. As a principle, it affirms the legal equality of States and guarantees their autonomy to determine their own internal and external affairs (Besson, Reference Besson2011). While sovereignty is often associated with exclusive territorial control, its more profound significance lies in decision-making autonomy—the entitlement of States to freely choose their political, economic, cultural, and foreign policies (Nicaragua v United States of America, 1986). This autonomy is not merely descriptive; it is juridically protected through the principle of non-intervention, which functions as a legal firewall against coercive interference.
The prohibition on intervention is an established—though notoriously underdeveloped—principle of customary international law (Nicaragua v United States of America, 1986). The prohibition developed in the 19th and 20th centuries, with a particular focus on the interventions in the Latin Americas (Helal, Reference Helal2025, p. 18). One of the first formal treaty articulations of the principle was Article 8 of the 1933 Montevideo Convention, which provided that “[n]o State has the right to intervene in the internal or external affairs of another” (Montevideo Convention, 1933). The scope of this principle was expanded in the Additional Protocol Relative to Non-Intervention in 1936, which prohibited intervention “directly or indirectly, and for whatever reason” (Additional Protocol Relative to Non-Intervention, 1936). The Organization of American States Charter specifically addressed that the prohibition “prohibits not only armed force but also any other form of interference of attempted threat” (Organization of American States, 1948). Such an approach rejected not only armed force but also subtler forms of interference, establishing a broad and anticipatory understanding of the norm.
This expansive regional understanding would later influence global legal developments. Although the Charter (United Nations, 1945) did not expressly codify a general prohibition on intervention, its spirit is discernible in Article 2(4), which prohibits the threat or use of force against the territorial integrity or political independence of any State. Notably, Article 2(4) does not extend to notions of economic force, which later bilateral and multilateral treaties relating to the prohibition on non-intervention do.
The broader principle, encompassing non-forcible acts, was clarified through soft law instruments such as General Assembly Resolution 2131 (United Nations General Assembly, 1965) on the Inadmissibility of Intervention and the 1970 Friendly Relations Declaration, which stated in sweeping terms that “no state or group of states has the right to intervene, directly or indirectly, for any reason whatever, in the internal or external affairs of any other state” (United Nations General Assembly, 1970). These provisions included within their ambit “all other forms of interference or attempted threats against the personality of the State or against its political, economic and cultural elements” (United Nations General Assembly, 1970).
Despite the hortatory nature of these declarations, the International Court of Justice (ICJ) in its 1986 judgment in Nicaragua (Nicaragua v United States of America, 1986) confirmed that the core provisions of the Friendly Relations Declaration reflected customary international law. The Court held that the principle “involves the right of every sovereign State to conduct its affairs without outside interference” (Nicaragua v United States of America, 1986, p. 202). The judgment confirmed that a prohibited intervention requires an interference in a state’s internal or external affairs that is coercive in character, stating that a prohibited intervention must
be one bearing on matters in which each State is permitted, by the principle of State sovereignty, to decide freely. One of these is the choice of a political, economic, social and cultural system, and the formulation of foreign policy. Intervention is wrongful when it uses methods of coercion in regard to such choices, which must remain free ones (Nicaragua v United States of America, 1986, p. 207).
It went on to clarify that a prohibited intervention must be coercive in character, and must relate to matters that fall within a State’s reserved domain—that is, areas where a State is entitled by international law to decide freely, such as the choice of political, economic, social and cultural systems, and the conduct of foreign policy.
The Nicaragua judgment is frequently cited as the most authoritative doctrinal statement on prohibited intervention. However, it left two critical questions unresolved: first, the precise scope of the reserved domain, and second, the threshold and definition of coercion, especially where it does not involve the use of force. The concept of a reserved domain—sometimes referred to as the domaine reserve—remains central to understanding what is protected by the prohibition on intervention. Yet as most international lawyers would be aware, a breach of sovereignty is not nearly as catastrophic as a layperson may be led to believe. It happens regularly, and with little consequence so long as it does not involve an armed attack or a use of force. Coercion thus occurs along a spectrum, rather than being a binary construct. It is a difficult concept theoretically, and even more difficult as a practitioner to understand the thresholds.
So how broad is this reserved domain? And how is coercion to be measured, especially in the absence of overt violence or threats? One method is to go back to basics and to understand exactly what the prohibition is seeking to protect. As Rosalyn Higgins notes:
The term ‘intervention’ only has a meaning measured against the question ‘intervention against what?’ and the answer has to be ‘intervention against a state’s domestic jurisdiction’ - that is, intrusion upon that which is for a state alone (Higgins, Reference Higgins2009, p. 274).
Here, the ICJ has provided two examples of what constitutes “internal and external affairs.” Milanovic highlights that these affairs are referred to as the reserved domain (domaine réservé) in the literature—“the reserved domain is the object of prohibited intervention, while coercion is its instrument or method” (Milanovic, Reference Milanovic2023, p. 609). The reserved domain is still limited by international law, and what States can freely decide within these bounds (Milanovic, Reference Milanovic2023, p. 609). However, the scope of the reserved domain remains broad, reflecting the principle (and rule) of sovereignty in international law (Milanovic, Reference Milanovic2023, p. 609).
It is clear that the prohibition protects again interferences with “the decisional sovereignty of a state in all its facets and manifestations in the absence of international obligations” (Roscini, Reference Roscini2024, pp. 165–6). Any operation, including a cyber operation, “intended to compel another State into compliance with its international legal obligations are excluded from the scope of applicability of the prohibition” (Schmitt, Reference Schmitt2017, p. 315).
These questions are of heightened importance in an era marked by transnational information operations, economic interdependence, and international regulatory regimes. In many cases, the scope of domestic discretion is itself contested, shaped by treaty commitments, international institutions, and evolving norms. For example, in some contexts, the exercise of sovereignty may be constrained by human rights obligations or trade rules. Intervention aimed at compelling compliance with such obligations may not, strictly speaking, violate the principle of non-intervention.
However, the legal situation becomes more complex in the domain of cyber operations and AI-enabled influence. These forms of interference frequently target populations directly—particularly through disinformation, behavioural nudging, or algorithmic amplification of certain narratives. As Townley notes, this raises a significant interpretive problem: does the principle of non-intervention protect the State as an institutional actor, or the people whose collective decisions give rise to State policy? (Townley, Reference Townley2019, p. 1168). The ICJ in Nicaragua assumed the former, stating that intervention is wrongful when it deprives the State of its free choice (Nicaragua v United States of America, 1986, p. 202; Tay, Reference Tay2022). Yet in democratic systems, public opinion and electoral processes are the mechanisms by which sovereign will is expressed. Operations that manipulate voters’ perceptions—without directly targeting State institutions—may still achieve the same coercive outcome (Malzac, Reference Malzac2022, p. 2). This is clearly a breach of sovereignty, as well as a probable breach of the prohibition on intervention.
The implication is clear: as new technologies empower States and non-State actors to shape cognitive and informational environments across borders, the existing doctrinal test for prohibited intervention is increasingly strained. A narrow reading of coercion, limited to physical or institutional pressure, fails to account for the indirect but potentially decisive ways in which AI, algorithms, and synthetic media can subvert sovereign autonomy.
In the sections that follow, this article argues that the principle of non-intervention must be recalibrated to reflect a broader, effects-based understanding of coercion, one which captures not only territorial incursions or economic pressure, but also systemic efforts to distort decision-making processes within the reserved domain. In doing so, it suggests that sovereignty today is as much about informational integrity as it is about territorial integrity—and that international law must evolve accordingly.
3. Digital sovereignty under threat: AI and the new logic of influence
The principle of sovereignty under international law has always served a dual function. On the one hand, it affirms a State’s control over its territory and population. On the other hand, and more fundamentally, it guarantees a State’s right to freely determine its internal and external affairs—what the International Court of Justice in Nicaragua referred to as the freedom to “conduct its affairs without outside interference.” In this sense, sovereignty is not merely spatial; it is decisional. Yet as States increasingly operate within a digitally mediated world, it is this latter dimension—sovereignty as autonomous decision-making—that is most under threat.
Emerging technologies, particularly AI, have accelerated a shift away from traditional forms of intervention and coercion. However, an important distinction must be made early. AI has transformed the scale, speed, and precision with which political influence can be exercised. Domestic lobby groups, political parties, and corporations can now deploy automated systems to distort debate, suppress turnout, or pressure decision-makers. Such practices may violate domestic law and international human rights obligations, but they do not in themselves constitute prohibited intervention. The legal distinction is structural. The prohibition on intervention protects States against external coercion in matters within their domaine réservé. When AI is deployed by domestic actors alone, the harm is internal. When the same techniques are covertly directed or enabled by a foreign State, the injury is different in kind: the displacement of sovereign authority by an outside power.
This distinction underpins the Decisional Integrity Model. Manipulation becomes unlawful not because it is technologically sophisticated, but because—when attributable to a foreign State—it substitutes for, or systematically degrades, another State’s capacity to determine its own political course.
Nor is AI-enabled intervention confined to elections or information campaigns. Autonomous botnets can overwhelm telecommunications networks; machine-learning systems can exploit vulnerabilities in electricity-grid management or transport controls; automated tools can disrupt emergency services. In such cases, the aim is not persuasion, but the neutralisation of the material conditions under which governmental authority is exercised. Where such operations are attributable to a foreign State and intentionally or foreseeably deprive a government of practical decisional capacity, they constitute a particularly direct form of coercive intervention—even if they fall short of armed force. By contrast, transient cyber incidents, isolated technical failures, and robust public diplomacy remain outside the scope of the prohibition. What matters is whether AI is used to replace or neutralise sovereign decision-making itself, rather than merely to compete within the marketplace of ideas.
No longer is influence exerted solely through force, threats, or direct economic pressure. Instead, States now possess tools capable of shaping another State’s perception of reality, manipulating electoral behaviour, controlling the flow of information, and altering how foreign governments and populations respond to political events. These technologies mark a fundamental transformation in the logic of international influence, and AI sits at the heart of it.
3.1 The amplification of influence
AI has not invented the challenge of coercive influence, but it has amplified it exponentially. As Jack Kenny observes, “[t]he use of sophisticated technological methods such as advanced AI techniques in influence operations do not change or alter the criteria on which coercion is evaluated,” but they “amplify the scale of effects that are possible to achieve from indirect interference operations” (Kenny, Reference Kenny, Cristiano, Broeders, Delerue, Douzet and Géry2023, p. 240). This amplification—of speed, reach, precision, and psychological impact—enables States to affect the internal affairs of others at a level of granularity and immersion previously unimaginable.
Through techniques such as deepfakes, algorithmic amplification, bot swarms, and microtargeted disinformation, AI allows foreign actors to simulate public consensus, distort public opinion, and erode trust in domestic institutions. Kenny details how “[a]rtificial intelligence technology allows actors to create numerous accounts, using machine learning to build fake user profiles and profile photos from user data,” which can then be deployed en masse to simulate a political movement. These accounts are used “to create a ‘herd mentality’ on online platforms during an election cycle,” thereby enabling “foreign actors to have a far more pervasive effect in influencing voter behaviour than if these profiles and their activity had to be created and maintained manually” (Kenny, Reference Kenny, Cristiano, Broeders, Delerue, Douzet and Géry2023, p. 235).
The strategic objective of such operations is not simply to inform or persuade but to displace a target State’s sovereign control over its own political processes. As Kenny explains,
[t]he challenge posed by such technology is whether its use in disinformation and microtargeting, if carried out extensively by the perpetrating state, could through influence alone amount to coercion in undermining the target state’s ability to conduct a free and fair election.
In one hypothetical scenario, Kenny imagines a deepfake video of an incumbent head of State announcing that an election is cancelled or that they have withdrawn from the race. If released on the eve of voting, and circulated virally, this misinformation could result in “a significant number of voters not casting their vote.” While such a scenario involves no attack on voting infrastructure, it would nonetheless “produce the same effect”: the distortion of electoral outcome through deception. “Such a scenario,” Kenny argues, “would likely amount to coercion and constitute a prohibited intervention” (Kenny, Reference Kenny, Cristiano, Broeders, Delerue, Douzet and Géry2023, p. 237).
3.2 The erosion of the reserved domain
Under international law, the prohibition on intervention protects a State’s reserved domain—those matters that international law permits a State to decide freely, including political organisation, foreign policy, and social systems. Coercion, in this framework, is the instrument that unlawfully intrudes upon these sovereign decisions. As Milanovic notes, “the reserved domain is the object of prohibited intervention, while coercion is its instrument or method” (Milanovic, Reference Milanovic2023, p. 609). AI technologies, while not altering this formal distinction, subvert its substance by rendering coercion more indirect, less attributable, but no less effective.
Influence operations conducted at scale by foreign actors “may push the boundaries of the interpretations of coercion,” Kenny writes, “particularly where they create an impression of a narrative among an electorate on online platforms that discredits a candidate or political party” (Kenny, Reference Kenny, Cristiano, Broeders, Delerue, Douzet and Géry2023, p. 236). Even where a State has the formal right to conduct an election, its ability to do so meaningfully may be hollowed out through technologically induced confusion, fatigue, or manipulation.
The legal challenge is that traditional understandings of coercion emphasise the voluntariness of the target State’s decision. Influence, it is often argued, does not deprive a state of agency, but merely seeks to persuade. However, Kenny challenges this view: “the fact that an operation seeks to compel an outcome through influence alone does not necessarily preclude it from compelling a state to act in an involuntary manner.” If a State is “deprived of control over its ability to hold a free and fair election,” he argues, that outcome is coercive (Kenny, Reference Kenny, Cristiano, Broeders, Delerue, Douzet and Géry2023, p. 236).
This logic applies well beyond elections. The systematic manipulation of a population’s informational environment—even without targeting official infrastructure—can nullify the meaningful exercise of sovereignty. Whether by confusing voters, discrediting public health messages, or triggering civil unrest, such operations constitute a strategic form of intervention whose effects are no less consequential for being bloodless.
3.3 Surveillance, data extraction, and the sovereign episteme
The threats posed by AI are not limited to psychological manipulation or electoral distortion. A second, equally significant domain of interference is the covert extraction and processing of data from within a State’s borders. Maruša Veber observes that the “non-consensual provision of humanitarian assistance with the use of AI” may involve the “unauthorised, clandestine gathering of large amounts of data on the territory of a concerned State” (Veber, Reference Veber2024, p. 241). Historically, such clandestine activity has been considered a violation of the principles of sovereignty and non-intervention. In one illustrative case, the Canadian Federal Court refused to authorise foreign surveillance under domestic law because it would involve activity “inconsistent with and likely to breach the binding customary principles of territorial sovereign equality and non-intervention” (Buchan, Reference Buchan2018, p. 52).
AI dramatically increases the scale and subtlety with which such surveillance can occur. Jaemin Lee notes that AI systems “can produce and process large quantities of data at a higher speed with greater accuracy,” making it easier for States to extract information from foreign populations and infrastructure without detection (Lee, Reference Lee2022, p. 147). This capability, he warns, is obscuring the boundaries of data sovereignty and “raising new challenges to be addressed by the international community” (Lee, Reference Lee2022, p. 147). The creation of economic or political value from such data, as in the case of Cambridge Analytica, illustrates how illicit data extraction can be weaponised to interfere in domestic politics. In such cases, “exploiting and sharing new economic values from illegally collected data could constitute an infringement of sovereignty” (Lee, Reference Lee2022, p. 152).
The legal architecture of the digital age has yet to catch up. As Lee notes, “just as AI makes it easier to collect data from individuals, AI implies easier and broader monitoring of other countries.” He warns that “autonomous artificial super-intelligence is difficult to track and control,” and that this raises the spectre of international disputes concerning the infringement of national sovereignty (Lee, Reference Lee2022, p. 153). Without clearer international regulation, there is a growing risk that sovereignty will be continually breached through silent, invisible incursions—not with tanks or sanctions, but with servers, algorithms, and behavioural datasets.
3.4 The doctrinal imperative
In sum, AI has not created a new legal crisis, but it has exacerbated an existing one. The line between permissible influence and prohibited coercion—already blurry—is now increasingly untenable. The principle of non-intervention, as currently framed, is ill-suited to detect and respond to informational interference that operates without overt threats or visible force. Yet if international law is to remain responsive to the purpose for which the non-intervention principle was designed—the preservation of sovereign autonomy—it must evolve.
What is at stake is not simply control over territory or formal governance. What is being contested is the sovereign episteme—the cognitive and informational space in which sovereign decisions are made. In the age of AI, protecting sovereignty requires protecting the conditions under which free and informed decisions can be made. This means defending not just institutions, but the informational integrity of populations. A legal framework that fails to address this is not merely incomplete—it is obsolete.
4. Coercion reimagined: legal tests for AI-enabled intervention
It does nothing for the project of international law, and its cornerstone documents, to simply bemoan new technologies’ growth and friction. As Rosalyn Higgins challenges:
And the task of the international lawyer over the next few years is surely not to go on repeating the rhetoric of dead events which no longer accord with reality, but to try to assist the political leaders to identify what is the new consensus about acceptable and unacceptable levels of intrusion (Higgins, Reference Higgins2009, p. 283).
The emergence of AI as a tool of statecraft has fundamentally altered the terrain of coercion and non-intervention in international law. Where earlier debates focused on threats, sanctions, or direct military assistance, today’s challenges arise from subtler forms of influence—algorithmically tailored persuasion, automated opinion shaping, and synthetic information. These AI-enabled operations often operate below the threshold of traditional force, yet they nonetheless interfere in the sovereign affairs of target States in ways that are predictable, scalable, and strategic. This section argues that the existing legal test for “coercion” is insufficient in the face of such developments and proposes a recalibrated model of coercion more consistent with the protective function of sovereignty under international law.
Despite its central role in delimiting unlawful intervention, the concept of coercion remains one of the least settled elements of the non-intervention principle (Milanovic, Reference Milanovic2023).
The notion of “coercion” under international law has historically been interpreted through two main lenses: intent-based and effect-based models. Under the intent-based model, a breach occurs only if the intervening State deliberately seeks to compel or constrain the target’s sovereign choices. Under the effect-based model, however, it is the result of the conduct—whether the target was effectively deprived of its decision-making freedom—that matters most, regardless of the intervener’s subjective purpose (Milanovic, Reference Milanovic2023, p. 626).
Neither model is entirely satisfactory when applied to AI-enabled interference. Intent is often difficult to prove in an environment where machine learning systems act semi-autonomously, and attribution remains elusive. Yet an effect-based standard risks overreach unless carefully delimited—after all, influence per se is not unlawful. The challenge, then, is to construct a nuanced model that avoids both over-and under-inclusiveness: one that can distinguish between unlawful coercion and lawful persuasion, and that reflects the qualitative change wrought by AI technologies.
The Tallinn Manual (Tallinn) offers some guidance in reconciling this dilemma, particularly as it moves away from a force-based understanding of coercion. As the Manual opines:
coercion must be distinguished from persuasion, criticism, public diplomacy, propaganda … retribution, mere maliciousness, and the like in the sense that, unlike coercion, such activities merely involve either influencing … the voluntary actions of the target State, or seek no action on the part of the target State at all (Schmitt, Reference Schmitt2017, p. 318).
Moreover, it suggests that coercion may arise even when the victim State retains formal choice, but where the operation in question “substantially deprives [it] of its ability to make decisions freely” reflecting the inherent link between sovereignty and prohibited intervention.
Experts contributing to the Tallinn process were divided on whether intent is necessary, but a majority accepted that foreseeability of effects may suffice in some contexts (Schmitt, Reference Schmitt2017, p. 315). This aligns with the observation that digital operations, while difficult to trace or impute, may nonetheless reliably produce political effects—such as public confusion, elite division, or international reputational damage—that diminish the autonomy of the State’s policy apparatus. The Tallinn Manual further holds that “[c]oercion is present when a state is compelled to act or abstain from acting in a matter within its sovereign prerogatives, regardless of the mechanism used” (Schmitt, Reference Schmitt2017, p. 315). This formulation is critical: it suggests that coercion is defined by reference to outcome, not by means.
Jack Kenny’s analysis of AI-enhanced disinformation echoes this conclusion. He argues that while “the use of sophisticated technological methods such as advanced artificial intelligence techniques in influence operations do not change or alter the criteria on which coercion is evaluated,” they “amplify the scale of effects that are possible to achieve from indirect interference operations” (Kenny, Reference Kenny, Cristiano, Broeders, Delerue, Douzet and Géry2023, pp. 224–25). In particular, Kenny notes that the ability of AI to generate “bot-based herd mentalities” or fabricate deepfake videos of political leaders can “deprive the target state of control over its ability to conduct a free and fair election to decide its government” (Kenny, Reference Kenny, Cristiano, Broeders, Delerue, Douzet and Géry2023, p. 237).
This is not merely theoretical. Consider a scenario in which a foreign State releases, via AI-generated deepfake, a video of a presidential candidate conceding defeat on the eve of voting. The video circulates virally and results in voter suppression or misdirection. Even though no voting infrastructure is compromised, the effect is functionally identical to tampering with the results: the will of the people is distorted through deception. Such operations, Kenny argues, “would likely amount to coercion and constitute a prohibited intervention” (Kenny, Reference Kenny, Cristiano, Broeders, Delerue, Douzet and Géry2023, p. 237).
4.1 New model: “decisional integrity”
At stake here is a more foundational issue: what is sovereignty, and what does it protect? While sovereignty traditionally connoted territorial integrity and legal independence, its normative core lies in decisional autonomy—the ability of a State to form and implement its own policies, free from external imposition.
AI operations strike precisely at this nerve. They do not coerce by threat or force, but by engineering the informational environment in which sovereign decisions are made. They alter not the choices themselves, but the conditions under which choices are formed. As Jaemin Lee notes, “[b]reaches of data collection and process standards among States without consent or notice via autonomic software and algorithms can be the subject of international disputes” (Lee, Reference Lee2022, p. 147). The boundary between influence and coercion is thus crossed not at the point of formal compulsion, but at the point where a State can no longer trust the cognitive substrate of its policy process.
This includes the population. While Nicaragua and related jurisprudence have framed the “State” as the direct object of coercion, AI-enabled operations that influence populations—particularly electoral constituencies—may, in effect, deprive the State itself of coherent agency. Indeed, the use of advanced AI techniques to create an impression of a narrative among an electorate pushes the boundaries of the interpretations of coercion. If a population is induced to act based on false premises intentionally constructed by a foreign State, then the resultant government decisions are no longer sovereign in any meaningful sense.
In response, this article proposes a recalibrated standard for coercion aligned with a modern account of decisional sovereignty: the Decisional Integrity Model. The model does not claim to displace existing intent-based or effects-based approaches, nor to introduce a wholly new category of wrongful intervention. Leading commentators have already sketched hybrid formulations that combine elements of purpose, means, and consequences in assessing coercion in the digital environment. Rather, the contribution of this model lies in its structured synthesis and operationalisation of those approaches into a decision procedure capable of application to AI-enabled influence operations.
Specifically, the model introduces two elements that are largely absent from prior tests. The first is a requirement of systemic effect: that the operation materially degrade the institutional, informational, or functional conditions under which a protected sovereign decision is formed, rather than merely influencing individual preferences within the ordinary contest of ideas. The second is the concept of functional substitution, capturing situations in which digital or AI-mediated processes effectively replace, disable, or commandeer governmental decision-making functions that international law reserves to the State itself. Together, these elements are intended to supply greater precision to the coercion inquiry, reduce false positives in cases of lawful persuasion or public diplomacy, and render the prohibition on intervention workable in environments characterised by automation, scale, and informational opacity.
On this account, coercion is defined as conduct attributable to a State that intentionally, or with foreseeability as to its consequences, produces a material and non-trivial impairment of another State’s capacity to exercise autonomous decision-making within its reserved domain, whether by systematically distorting the decisional environment or by functionally supplanting core governmental processes.
To demonstrate both the practical operation and the limits of the Decisional Integrity Model, the analysis that follows applies the framework to a series of worked hypotheticals extending beyond electoral interference. These include scenarios involving AI-assisted data exfiltration and psychographic micro-targeting affecting public health decision-making, algorithmic amplification of secessionist narratives culminating in civil unrest, and large-scale data poisoning of government datasets used for regulatory purposes. Each case is examined element by element, including instances in which the model should not be satisfied, in order to illustrate threshold effects, evidentiary constraints, and the distinction between unlawful coercion and lawful—even if aggressive—persuasion. This calibrated approach is intended to guard against over-extension of the non-intervention principle while preserving its capacity to address genuinely coercive forms of digital and AI-enabled interference.
This model incorporates five elements:
-
1. Attribution—The act must be imputable to a State under Article 4 or 8 of the ILC’s Articles on State Responsibility (International Law Commission, 2001).
-
2. Target Domain—The act must relate to a matter falling within the target State’s reserved domain.
-
3. Predictable Deprivation—The act must foreseeably impair the target State’s ability to make a sovereign decision free from external manipulation.
-
4. Systemic Effect—The impairment must affect not merely one official or policy choice, but the conditions for institutional decision-making (e.g. electoral integrity, public information environment, elite consensus).
-
5. Functional Substitution—The foreign State must effectively substitute its own preferences for those of the target State—either by misdirection, suppression, or falsification of the target’s policy output.
Each are discussed in more detail below.
4.1.1 Attribution—linking the act to a state
The first condition for determining whether an act constitutes coercion under international law is attribution. Attribution ensures that the conduct in question can be legally imputed to a State, thus engaging that State’s international responsibility. As codified in Articles 4 and 8 of the International Law Commission’s Articles on State Responsibility (ASR), this requirement distinguishes acts that represent the conduct of a State from those that are merely private or non-State in character. Article 4 provides that the conduct of any State organ—regardless of whether it exercises legislative, executive, judicial, or other functions—is considered an act of the State under international law (International Law Commission, 2001). Article 8 extends this principle by imputing responsibility for the conduct of non-State actors when such conduct is carried out “on the instructions of, or under the direction or control of,” a State (International Law Commission, 2001).
This element becomes particularly salient in the context of AI-enabled interference operations, which are often conducted through proxies, outsourced digital influence networks, or algorithmically generated content farms. Unlike conventional forms of military or diplomatic coercion, AI-driven influence campaigns are designed for plausible deniability. A State may support or direct a campaign of deepfakes, microtargeted propaganda, or coordinated inauthentic behaviour without leaving obvious fingerprints. Yet if the operational control, financial backing, or strategic coordination of such an operation can be linked to a government authority—or if a pattern of practice reveals deliberate orchestration—it satisfies the attribution threshold under international law.
As technologies become more autonomous, this requirement will not become obsolete, but more complex. For example, if a State deploys a machine-learning system to target a foreign electorate, and the system evolves novel tactics of manipulation over time, the original deploying state remains responsible if the system was launched and trained for that purpose. In such cases, attribution must encompass both direct and indirect control, as well as the foreseeability of outcomes from semi-autonomous digital tools. The attribution requirement thereby guards against overreach, but it also ensures that States remain accountable for their informational proxies.
4.1.2 Target domain—interference in the reserved sovereign sphere
The second element, target domain, requires that the act in question interfere with matters that fall within a State’s reserved domain or domaine réservé—those areas of activity that international law explicitly leaves to a State’s own discretion. The ICJ in Nicaragua described this as including a State’s choice of “political, economic, social and cultural system,” as well as the formulation of foreign policy (Nicaragua v United States of America, 1986, p. 202). These domains represent the heartland of sovereignty: the foundational power of a State to determine its internal and external affairs free from outside coercion.
The principle of non-intervention is built on the idea that international law protects a State’s ability to make autonomous decisions in these areas. While not all forms of influence infringe this sphere—public diplomacy or mutual treaty negotiation are expected elements of international life—acts that aim to override or circumvent these choices engage the legal protections of the reserved domain. When AI-enabled tools are deployed to manipulate electoral processes, fabricate public sentiment, or distort national security debates, they often strike at these very domains.
Importantly, the reserved domain is not static. As the means by which sovereign choices are made evolve—from physical ballots to digital platforms, from public speeches to algorithmic recommendation engines—so too must the legal understanding of what constitutes protected sovereign space. The domain now includes not only traditional forums such as legislatures or cabinet deliberations, but also the digital and informational conditions under which political will is formed. Efforts to hijack, flood, or algorithmically distort this space represent a modern form of interference. In this sense, the target domain element ensures the principle of sovereignty remains fit for purpose in an era of algorithmic statecraft.
4.1.3 Predictable deprivation—foreseeable impairment of sovereign will
The third component of the Decisional Integrity Model introduces the notion of predictable deprivation. It asks whether the foreign conduct foreseeably impaired the target State’s capacity to make sovereign decisions free from external manipulation. This requirement borrows from principles of due diligence and foreseeability found in other areas of international law—such as environmental law, the law of armed conflict, and international human rights law—where responsibility may attach not only to intentional harm, but also to negligent or reckless conduct where harmful effects were reasonably foreseeable.
AI systems introduce new dynamics to this inquiry. Modern influence operations are not accidental; they are designed using data analytics, behavioural profiling, and iterative machine-learning techniques that enable real-time feedback and optimisation. When a State deploys such tools to sway electoral outcomes, mislead public officials, or destabilise trust in media institutions, it cannot plausibly deny the foreseeable nature of these effects. For example, if a synthetic video of a political candidate conceding an election is released on the eve of voting, and this results in widespread voter confusion or abstention, the producing State cannot claim ignorance of the impact.
What matters here is not merely the presence of intent, but the quality of the harm that results. Did the conduct foreseeably impair the target State’s ability to make a sovereign decision in its own interest, on its own terms, and in its own voice? If so, then predictable deprivation is established. This standard is tailored to the digital era, where States may not issue threats or deploy troops, but can nonetheless predict—and even model—the erosion of another State’s decisional autonomy.
4.1.4 Systemic effect—undermining the infrastructure of decision-making
A crucial innovation of this model is its inclusion of systemic effect as a criterion for unlawful coercion. This requirement demands that the impairing effect of the foreign conduct be felt not just in a single decision or by a lone actor, but across the institutions and practices that constitute a State’s capacity to decide. These include electoral systems, media environments, judicial independence, civil-military relations, public education, and elite policy formation processes. Sovereignty is not merely the freedom to make one decision—it is the structural integrity of decision-making itself.
AI tools are uniquely capable of affecting this dimension. By flooding social media with coordinated messages, targeting journalists and opposition leaders with synthetic content, or corrupting public datasets, a foreign actor can warp the epistemic environment within which policy choices are made. These actions do not need to be dramatic to be effective. In fact, their subtlety is often their strength: they are harder to detect, less likely to provoke retaliation, and capable of spreading mistrust or confusion across multiple levels of governance. As Jaemin Lee observes, “just as AI makes it easier to collect data from individuals, AI implies easier and broader monitoring of other countries,” enabling forms of surveillance and distortion that bypass physical intrusion (Lee, Reference Lee2022, p. 153).
To qualify as a systemic effect, the coercive conduct must impair the conditions under which decisions are made—by altering perceptions, fracturing consensus, or disabling key intermediaries. This element draws attention to sovereignty as a process, not merely an outcome. It ensures that legal protection attaches not just to the moment of formal decision-making, but to the institutional ecology that makes such decisions legitimate and possible.
To guard against overbreadth, this element is subject to threshold requirements. The impairment must be material rather than speculative, demonstrable through objective indicators such as measurable degradation of institutional capacity, significant and attributable distortion of electoral participation or outcomes, sustained disruption of information infrastructures relied upon by public authorities, or quantifiable erosion of the functioning of core governance systems. Isolated misinformation, transient controversy, or marginal shifts in public sentiment are insufficient.
Scale and duration are likewise relevant. Systemic effect ordinarily requires operations of sufficient magnitude, persistence, or geographic or demographic reach to plausibly affect institutional functioning, rather than episodic or low-level information activity. A proximity filter further applies: the operation must bear a close and identifiable connection to a specific sovereign decision or decision-making process within the target State’s domaine réservé (such as an election, public health emergency response, territorial determination, or regulatory choice), and not merely contribute to the general information environment in an indirect or diffuse manner.
Finally, reversibility and the availability of effective countermeasures are relevant to calibration. Where domestic institutions, independent media, platform governance mechanisms, or technical mitigation measures are realistically capable of neutralising the effects of the operation in a timely manner, the threshold of systemic impairment will ordinarily not be met.
4.1.5 Functional substitution—replacing the target state’s will
The final element, functional substitution, is the clearest indicator that coercion has occurred. It refers to the effective replacement of the target State’s preferences with those of the intervening State—whether through misdirection, suppression, or falsification of decision outputs. While earlier models of coercion often required physical control or explicit threats, functional substitution captures subtler but no less damaging tactics: artificially manufacturing political mandates, distorting elite perceptions, or hijacking public consensus to deliver an outcome that reflects the will of a foreign power.
Functional substitution operates by crowding out or overriding sovereign will. For instance, if bots controlled by a foreign State flood an online platform with false information about a referendum, leading to a vote that would not otherwise have occurred or that misrepresents the true preferences of the people, then the foreign State has substituted its preferences into the sovereign process. Similarly, if machine-learning algorithms are deployed to create the appearance of a grassroots movement, thereby pressuring officials to adopt policies contrary to national interest, this too qualifies.
This element links the concept of coercion to the traditional concern of domination in international relations—a definition that fits precisely where foreign actors use AI not to persuade, but to simulate, displace, or impersonate the policy preferences of the target State. Functional substitution is the final test of the model because it marks the point at which influence ceases to be advisory and becomes decisional. It is not merely an intrusion into another State’s processes—it is an appropriation of those processes, by stealth or surrogacy.
Such a model balances the need to prevent abuse with the obligation to protect meaningful sovereignty in a digital age. It does not criminalise influence per se—public diplomacy, cultural outreach, and bilateral persuasion remain lawful. But it captures the quantitative and qualitative shift wrought by AI, recognising that modern intervention often occurs not through domination of territory or institutions, but through domination of perception.
For avoidance of doubt, functional substitution does not encompass truthful, attributable expression by States, even where sharply critical or strategically timed, nor ordinary diplomatic advocacy, public messaging, or support for civil society actors. These remain within the domain of lawful persuasion. By contrast, deception, impersonation of domestic actors or institutions, covert manipulation of information architectures, inauthentic amplification at scale, or the exploitation of unlawfully obtained personal or governmental data constitute strong indicia of substitution rather than persuasion, particularly where they are tightly linked to a protected sovereign decision.
Accordingly, functional substitution requires more than influence; it requires a demonstrable displacement of the target State’s decisional agency in relation to a specific governmental choice. Where such displacement cannot be shown, even extensive foreign messaging or criticism will not satisfy the coercion threshold under this model.
5. Conclusion
Sovereignty has long been defined by borders—by the capacity of a State to control territory, govern its population, and resist external interference. But in the algorithmic era, these classical contours are no longer sufficient. As this article has demonstrated, emerging technologies such as AI, synthetic media, and behavioural analytics permit a qualitatively new mode of intervention: one that targets not the State’s physical integrity, but the informational conditions under which it thinks, decides, and acts.
In response, this article has proposed a recalibrated framework—the Decisional Integrity Model—that adapts the legal test for coercion to the realities of digital interference. By grounding coercion in the predictable degradation of a State’s decision-making autonomy, and linking it to systemic institutional effects and functional substitution, this model brings coherence to a field long constrained by ambiguity. It offers a principled basis for distinguishing lawful persuasion from unlawful manipulation, and does so without chilling legitimate diplomacy or political expression.
Yet if the law is to keep pace with technological transformation, the principle of non-intervention itself must be updated. The next phase of doctrinal development must recognise that sovereignty is no longer exercised solely through territory, but also through cognition. States must be entitled to informational self-determination—the ability to form preferences, evaluate evidence, and choose freely without being deceived, overwhelmed, or pre-empted by foreign algorithmic systems. This does not imply a right to epistemic insulation or immunity from criticism. But it does demand protection from foreign conduct that functionally replaces a State’s political will with another’s.
Objections will arise. Free expression advocates may argue that regulating digital influence risks suppressing legitimate speech. But this concern can be addressed by distinguishing expression from manipulation. Attribution challenges—always present in the cyber domain—must be met with robust evidentiary standards, not abandonment of legal accountability. And those who invoke the neutrality of technology must be reminded that law has always regulated means, not just ends: a tool does not become exempt from scrutiny merely because it is efficient.
The implications of this reconceptualisation are far-reaching. For legal theory, it demands that sovereignty be reconceived not as a fixed entitlement, but as a capacity: the ongoing ability to decide. For diplomatic practice, it compels a recalibration of acceptable conduct in the digital sphere, including clearer norms against electoral interference, data theft, and algorithmic distortion. For international law, it offers a foundation for future treaties and soft law instruments aimed at protecting the epistemic integrity of States.
As international law once evolved to respond to airspace, nuclear weapons, and cyberspace, so too must it now evolve to meet the challenge of AI. The Decisional Integrity Model is a first step towards that goal. It defends not only the sovereign equality of States, but their ability to remain the authors of their own futures—even in a world where perception itself is increasingly programmable.
Open access
