Ch. 14: Digital Sovereignty and the Governance of Legal Infrastructure

The Sovereignty Question in the Age of Executable Law

The question of sovereignty has returned to European governance in a new form. It no longer appears only in the traditional language of territory, defence, or industrial capacity. Increasingly, it appears in the language of infrastructure. Who controls the systems on which public authority depends? Who operates them? Who can inspect them? Who can change them and who can withdraw them?¹ In the digital state, these are no longer technical questions at the edge of constitutional life. They are among its central questions.

This is the context in which digital sovereignty must be understood. It is not a slogan for local hosting, nor a rhetorical preference for domestic suppliers. It is the capacity of a political community to exercise public authority through infrastructures that remain genuinely subject to its own law, institutions, and democratic control. A state that depends on digital systems it cannot inspect, cannot govern, and cannot replace on its own terms has not simply acquired a technical dependency. It has created a constitutional vulnerability at the very point where law becomes administration, and administration becomes lived reality for the citizen².

That vulnerability becomes especially acute in the domain of legal administration. The machine-executable form of a norm is not peripheral to sovereign authority. It is one of its operative expressions. It is the point at which a legislature’s abstract command is translated into the concrete determination of whether a benefit is granted, a licence issued, a tax assessed, or an enforcement measure triggered. Once the application of law depends on executable artefacts, the infrastructure that stores, publishes, versions, authenticates, and serves those artefacts becomes part of the machinery of sovereignty itself. To lose effective control over that layer is not merely to outsource IT. It is to place the operational form of law in a position of dependency.

This is why the OLRF treats legal infrastructure as sovereign infrastructure. That claim is not an ornament added for strategic effect. It follows directly from the argument made earlier in this paper. If machine-executable law is a public utility, if it is part of the infrastructure through which the state secures legal certainty and gives effect to the social contract, then control over that infrastructure becomes a matter of constitutional importance. The decisive distinction is not between digital and analogue government. It is between a legal order whose operative infrastructure remains public, open, and institutionally governable, and one whose most consequential functions are quietly absorbed into private platforms.

The three models sharpen the sovereignty question in ways that a single-model architecture would not reveal. Under Model A, the sovereignty domain is the normative layer: the Decision Tree, the Coverage Map, the Registry, and the evaluation engine. These are the artefacts and systems over which the state must retain control. Under Model B, the sovereignty domain expands to include the agent infrastructure: the Legal Agent that subsumes facts under legal concepts is performing a normative function, and the question of who provides, trains, certifies, and governs that agent becomes sovereignty-relevant. A state that controls its Decision Trees but depends on a foreign-controlled AI service for the subsumption step has not secured its normative sovereignty. It has merely relocated the dependency from the tree to the agent. Under Model C, the sovereignty domain expands further: the autonomous agent that reasons directly from statutory text is, in functional terms, exercising a form of legal interpretation. If that agent is provided by a commercial actor outside democratic control, the state has outsourced not merely the application of law but a significant element of its interpretation³.

The Sovereignty Risk in the Platform Age

The difficulty is not hard to see. Over the past two decades, a small number of large technology firms have accumulated a combination of cloud scale, software ecosystems, AI capability, and operational maturity that makes them increasingly attractive partners for the public sector. Their offer is compelling. They can provide managed services, advanced tooling, rapid deployment, and integration across administrative workflows at a level that few governments can reproduce quickly on their own. The efficiency gains are real. The strategic risk lies elsewhere.

The risk is that convenience at the application layer grows into a dependency at the constitutional layer. If the executable form of law is specified, hosted, versioned, and served through infrastructure that is effectively controlled by external commercial actors, then the state’s legal autonomy becomes contingent on relationships it does not fully govern. Even where formal ownership remains public, practical dependence may still arise through proprietary formats, opaque extension layers, closed toolchains, exclusive service dependencies, and accumulated operational knowledge that cannot be transferred without prohibitive cost. What appears at first as procurement can become, over time, a loss of institutional freedom of action⁴.

The problem is not that commercial actors participate in public digital systems. They always will, and in many places they should. The problem is that in the absence of architectural safeguards, their participation can migrate from implementation and service provision into effective control over the legal infrastructure itself⁵. Once that happens, a public authority may still issue the norm in formal terms, but the practical life of that norm will depend on infrastructures, interfaces, and extension mechanisms that lie outside democratic control. The state remains sovereign in theory while becoming dependent in operation.

Under the three-model framework, this risk takes a specific and increasingly urgent form. The AI models that power Legal Agents under Models B and C are, at the time of writing, overwhelmingly provided by a small number of US-based and Chinese-based companies. A European state that adopts Model B or C for high-volume administrative processes, using agents powered by foundation models from these providers, has created a dependency that is qualitatively different from the cloud dependency that European sovereignty policy has already identified. It is not a dependency on compute or storage. It is a dependency on the reasoning infrastructure through which its law is applied. The sovereignty implications of this dependency have not yet been adequately addressed in European digital policy⁶.

Legal Infrastructure as a Distinct Strategic Domain

European digital sovereignty policy has, for understandable reasons, focused on cloud, data, semiconductors, identity, and AI capability. Those domains matter enormously. Yet legal infrastructure has often remained absent from the sovereignty conversation, as if the systems through which norms are specified, published, and executed were merely a subcategory of administrative software. That absence is increasingly untenable. As governance becomes more automated and as AI systems become more deeply integrated into administrative processes, legal infrastructure can no longer be treated as a secondary implementation detail. It is one of the infrastructures through which democratic self-government is either preserved or hollowed out.

This is why the OLRF should be understood as a contribution to the European sovereignty agenda, and not just to legal technology or public sector modernisation. A Europe that has invested in sovereign cloud capacity, in trusted identity, in federated data spaces, and in strategic AI capability, but that leaves the machine-executable form of its laws dependent on foreign-controlled platforms, has not solved the sovereignty problem at its most sensitive institutional point. It has modernised around the law while leaving the operative form of law itself exposed. That would be a strategic contradiction⁷.

The three-model framework shows that the strategic domain is not one but three. The first is the normative infrastructure: the Registry, the Decision Trees, the Coverage Maps, the DataPoint Schemas, the evaluation engine. This is the sovereignty domain of Model A, and it must be secured regardless of which model is adopted. The second is the agent infrastructure: the AI systems that perform subsumption (Model B) or autonomous reasoning (Model C), including the foundation models on which they are built, the training data from which they learn, and the certification systems through which they are assessed. The third is the coordination infrastructure: the protocols and registries through which agents discover each other, delegate tasks, and assemble composite outcomes across jurisdictions. Each domain requires its own sovereignty strategy. Securing the first without securing the second is insufficient for any state that intends to adopt Model B or C at scale.

Federation as the Institutional Form of Sovereignty

The OLRF’s answer to the normative sovereignty problem is not centralisation. It is federation. The federated Registry model is the institutional form through which sovereignty and interoperability are reconciled. Each jurisdiction maintains its own Registry instance, under its own law, with control over its own norms, signing keys, access policies, and governance procedures. Cross-border interaction does not require surrender of control to a central operator. It is achieved through standardised interfaces that allow one jurisdiction’s systems to query another’s Registry under defined conditions, without administrative access to the other jurisdiction’s internal infrastructure.

This design choice matters because sovereignty in Europe is not a unitary phenomenon. It is layered. National legal orders retain their own constitutional identities and administrative structures even where they operate within a larger European framework. Any architecture for executable law that required those orders to collapse into a single administrative centre would fail politically long before it failed technically. Federation avoids that trap. It preserves national control while creating the conditions for continental interoperability. In that sense, it mirrors the institutional logic of other successful European infrastructures, including cross-border identity and data exchange frameworks, while adapting that logic to the specific demands of executable norms⁸.

The handling of norm conflicts in a federated system illustrates the point. Where a cross-border case implicates the executable norms of more than one jurisdiction, the Registry does not decide which law prevails. It makes the conflict visible. It enables the creation of a cross-border decision record that shows how the same facts are evaluated under different Decision Trees and on what normative basis. The legal resolution of that conflict remains where it belongs: in private international law, treaty arrangements, and ultimately judicial determination. The infrastructure supports the visibility of law across borders. It does not usurp the authority to settle conflicts between them.

The Open Standard and the Risk of Enclosure

Yet federation alone is not enough. A formally open system may still be proprietary in practice. The history of digital infrastructure offers many examples of this pattern. A standard begins as open and widely accessible. A commercial actor adopts it, extends it, builds a market around the extended version, and gradually turns proprietary additions into the real conditions of participation. The original standard remains open on paper, but its practical significance erodes. Others may still implement it, yet they can no longer compete effectively without reproducing the private extensions that now define the ecosystem⁹.

In the normative domain, that risk would be of high significance. A Decision Tree format that remained nominally open while being functionally captured by proprietary evaluation semantics, anchor schemes, or Registry interfaces would reproduce the very dependency the OLRF is intended to prevent. Public authorities would appear to retain control over their legal artefacts, while in practice their portability, interoperability, and long-term autonomy would depend on vendor-specific layers. The standard would remain open in name and closed in consequence.

Under the three-model framework, the enclosure risk extends beyond the normative layer. A proprietary validation framework for Model B, or a proprietary audit protocol for Model C, would create the same kind of dependency at the agent governance level that a proprietary Decision Tree format would create at the normative level. The OLRF’s open standard commitment must therefore extend to the full scope of what the architecture publishes: not only Decision Trees and Coverage Maps, but also validation frameworks, deviation classification criteria, audit protocols, and agent certification standards. Any element that forms part of the constitutional infrastructure must remain open, portable, and vendor-independent.

Governance of the Commons

The OLRF’s answer is a layered governance model designed to keep the commons genuinely common. That model has three core components: copyleft licensing for the core specification, multi-stakeholder governance for its evolution, and extension neutrality as an architectural requirement for conformance. These are not incidental design features. They are the governance conditions under which an open standard can remain a sovereignty instrument over time.

Copyleft addresses the first danger. It ensures that extensions or derivative forms of the core specification, when distributed for production use, cannot be taken private while continuing to shape the public infrastructure on which others depend. Commercial actors remain free to build services, engines, tooling, and operational offerings around the open core. What they cannot do is alter the core specification and convert those alterations into proprietary control points that others must accept in order to remain interoperable. Competition is therefore directed toward implementation quality, service reliability, performance, and value-added capabilities, rather than toward private ownership of the normative substrate itself¹⁰.

Multi-stakeholder governance addresses the second danger, which is capture through institutional control rather than code. A sovereignty-preserving standard cannot be allowed to evolve solely according to the interests of its most powerful adopters. Public authorities, smaller jurisdictions, civil society, academia, and commercial actors must all have a place in its development, and in a form that prevents any one category of participant from shaping the standard to its own structural advantage. The governance procedures through which the standard evolves must be publicly answerable rather than commercially steered¹¹.

Extension neutrality addresses the third danger, namely that proprietary additions might become operationally indispensable even where the core remains formally open. The principle is simple. A conformant implementation of the core specification must remain independently functional and interoperable without requiring proprietary additions. A public authority that expresses its executable norms in the OLRF core should be able to move between conformant vendors, preserve access to its Registry history, and continue operating its legal artefacts without rewriting them into a platform-specific format. That is the technical expression of sovereign freedom of action. It means that the state’s normative layer is not held hostage by any single commercial relationship.

Commercial Participation Without Constitutional Dependency

None of this implies hostility to the market, as it just refuses the defective choice between public control and commercial participation. The OLRF does not seek to exclude private actors from the ecosystem. It seeks to structure their participation so that commercial dynamism strengthens the commons rather than enclosing it. There will be room for evaluation engines, Registry operations, integration services, explanation tooling, AI fact-finding components, compliance services, and high-quality managed offerings. The point is that these should compete on implementation and service, not on ownership of the legal substrate¹².

That distinction is fundamental. A healthy ecosystem is not one in which the public sector builds everything itself. It is one in which public authorities retain control over the normative core while commercial actors compete to serve that core better. Such a market is fully compatible with innovation, and indeed more likely to sustain it over time, because it avoids lock-in, lowers entry barriers for new participants, and protects the portability of the public artefacts on which the ecosystem depends. In this respect, sovereignty and competition are not opposites. Properly designed, they are mutually reinforcing and a core requirement for a tech driven API-first public infrastructure.

Under the three-model framework, the commercial ecosystem expands significantly. Model A creates a market for deterministic evaluation engines, Registry hosting, Decision Tree authoring tools, and Coverage Map generators. Model B creates an additional market for Legal Agent providers, validation framework implementations, and deviation analysis services. Model C creates a further market for autonomous reasoning agents, audit protocol implementations, and population-level outcome testing services. Each layer of the market operates on top of the open normative substrate. No layer requires proprietary control over that substrate in order to compete effectively. The commercial opportunity grows with the models. The constitutional dependency does not, provided that the open-standard commitment is maintained at every level.

Sovereignty by Design

The central claim of this chapter is therefore straightforward. In an age of executable law, sovereignty cannot be understood only in terms of who enacts the norm. It must also be understood in terms of who controls the infrastructure through which that norm is specified, published, authenticated, served, and applied. A state that loses control over this layer may retain formal legal authority while ceding the practical conditions under which authority takes effect.

The three-model framework makes this claim more precise and more urgent. Under Model A, sovereignty over the normative layer is necessary and, for many administrative processes, sufficient. Under Models B and C, sovereignty over the normative layer alone is insufficient. The state must also secure its capacity to govern the agent infrastructure through which its law is applied. That means the ability to certify, audit, and if necessary replace the AI systems that perform normative reasoning, without dependence on any single commercial provider. It means the ability to set and enforce standards for agent behaviour that reflect the constitutional values of the legal order the agent serves. And it means the ability to do all of this under the state’s own law, through the state’s own institutions, and subject to the state’s own democratic accountability¹³.

The OLRF responds by treating legal infrastructure as a constitutional object. Its federated Registry architecture protects jurisdictional control while enabling cross-border interoperability. Its open standards prevent dependence on private interfaces as the price of participation. Its governance model protects the commons against enclosure. And its three-model framework makes visible the escalating sovereignty requirements that accompany increasing AI autonomy, so that the choice between models is made with full awareness of its constitutional and strategic implications.

For that reason, the question posed by this chapter is not whether digital sovereignty matters for law. It is whether a legal order that becomes machine-executable can remain sovereign without designing that sovereignty into its infrastructure from the beginning. The answer offered by the OLRF is clear. It cannot. Sovereignty, in this domain, must be built into the architecture before dependency becomes the background condition of governance.

The certification system (Chapter 10) is the primary instrument through which the state exercises sovereignty over the agent layer. The state exercises sovereignty over the normative layer through the Registry: it publishes, versions, signs, and governs its own Decision Trees. It exercises sovereignty over the agent layer through certification: it defines the qualification requirements, administers the examinations, issues and revokes the credentials, and publishes the results. A state that publishes its own Decision Trees but accepts uncertified agents from foreign providers has secured normative sovereignty while ceding agent sovereignty. It controls what the law says in executable form, but not who applies it. Under Models B and C, where agents exercise normative functions, that gap is constitutionally untenable. Certification closes it. It ensures that every agent that applies a state’s law within that state’s jurisdiction has been assessed against that state’s standards, under that state’s institutional authority, and subject to that state’s revocation power. In this sense, the certification system is not an administrative appendage to the OLRF architecture. It is the sovereignty mechanism for the agent layer, as the Registry is the sovereignty mechanism for the normative layer.

Nemitz, P., “Constitutional Democracy and Technology in the Age of Artificial Intelligence”, Philosophical Transactions of the Royal Society A, Vol. 376, No. 2133, 2018, pp. 1 ff., arguing that the questions of who controls, who operates, and who can withdraw the technological systems on which public authority depends are not questions of procurement or IT governance but of constitutional law, because the practical capacity to exercise democratic self-government is contingent on the institutional governability of the infrastructure through which public power takes effect. Nemitz, who served as principal adviser in the European Commission’s Directorate-General for Justice, frames this as a democratic precondition: a state that cannot inspect or replace the systems through which it administers law has not merely accepted a technical dependency but has ceded a constitutional capacity ↩
The concept of digital sovereignty as the capacity of a political community to exercise public authority through infrastructures subject to its own law and democratic control draws on: Floridi, L., “The Fight for Digital Sovereignty: What It Is, and Why It Matters, Especially for the EU”, Philosophy and Technology, Vol. 33, 2020, pp. 369 ff.; Pohle, J. and Thiel, T., “Digital Sovereignty”, Internet Policy Review, Vol. 9, No. 4, 2020; European Parliament, Report on European Digital Sovereignty, A10-0107/2025. For the constitutional dimension: Hoffmann-Riem, W., “Verhaltenssteuerung durch Algorithmen: Eine Herausforderung für das Recht”, AöR, Bd. 142, 2017, S. 1 ff. ↩
The proposition that sovereignty over the agent infrastructure is constitutionally distinct from sovereignty over the normative infrastructure is, to our knowledge, novel in this form. Its logic follows from the principle that wherever an AI system performs a function that was previously reserved to a human official exercising public authority, the governance of that AI system is a matter of public law, not merely of procurement law. See: Martini, M. and Nink, D., “Wenn Maschinen entscheiden… Vollautomatisierte Verwaltungsverfahren und der Persönlichkeitsschutz”, NVwZ-Extra 10/2017, pp. 1 ff.; for the broader argument that AI governance in the public sector is a sovereignty question: Katzenbach, C. and Ulbricht, L., “Algorithmic Governance”, Internet Policy Review, Vol. 8, No. 4, 2019. ↩
The mechanism by which procurement convenience transforms into constitutional dependency is analysed in: Morozov, E., “Digital Socialism? The Calculation Debate in the Age of Big Data”, New Left Review, Vol. 116/117, 2019; for the specific European context: Celeste, E., “Digital Constitutionalism: A New Systematic Theorisation”, International Review of Law, Computers and Technology, Vol. 33, No. 1, 2019, pp. 76 ff. The analogy to energy dependency is developed in: Berlin Declaration on Digital Society and Value-Based Digital Government, December 2020, para. 14 ff. ↩
Rahman argues that private actors who control essential infrastructures exercise a form of power that is functionally equivalent to governmental authority, because their decisions about access, pricing, standards, and interoperability determine the practical conditions under which all other actors, including public authorities, must operate. The constitutional danger is not the participation of private actors per se but the absence of governance structures that prevent operational dependency from converting into normative authority. See: Rahman, K. S., “The New Utilities: Private Power, Social Infrastructure, and the Revival of the Public Utility Concept”, Cardozo Law Review, Vol. 39, 2018, pp. 1621 ff.; ↩
The concentration of foundation model provision among a small number of non-European actors is documented in: Stanford HAI, “Artificial Intelligence Index Report 2025”, Chapter 1 (industry concentration); European Commission, “2024 Report on the State of the Digital Decade”, COM(2024) 260 final, pp. 28 ff. (AI capability gap). The sovereignty implications for public administration specifically are analysed in: Ulbricht, L., “When Big Tech Provides the State’s AI: Democratic Risks and Regulatory Responses”, Government Information Quarterly, Vol. 41, No. 1, 2024. ↩
The characterisation of legal infrastructure as a distinct strategic domain within the broader European sovereignty agenda extends the argument of the Berlin Declaration (2020, op. cit.) and the European Commission’s “Shaping Europe’s Digital Future”, COM(2020) 67 final, which identifies sovereign digital infrastructure as essential to strategic autonomy but does not explicitly include the machine-executable form of law within its scope. ↩
The federated model draws on the design principles of Gaia-X (Gaia-X Architecture Document, Release 22.10, 2022) and the European Digital Identity Framework (Regulation (EU) 2024/1183, amending Regulation (EU) No 910/2014 on electronic identification and trust services, eIDAS 2.0). For the constitutional argument that federation is the appropriate institutional form for European digital infrastructure because it mirrors the layered sovereignty structure of the European legal order: Bogdandy, A. von, “Founding Principles”, in Bogdandy, A. von and Bast, J. (eds.), Principles of European Constitutional Law, Hart Publishing 2009, pp. 11 ff. ↩
The pattern of “embrace, extend, extinguish” is well documented in technology history: Simcoe, T. and Watson, J., “Forking, Fragmentation, and Splintering”, Strategy Science, Vol. 4, No. 4, 2019, pp. 283 ff. For the specific risk of enclosure in open standards for public infrastructure: Blind, K., “The Impact of Open Standards on the Competitiveness of European Industry”, European Commission DG Enterprise 2011. For the normative domain specifically: Russell, A., Open Standards and the Digital Age: History, Ideology and Networks, Cambridge University Press 2014. ↩
The use of copyleft licensing as a governance mechanism for public digital infrastructure, rather than merely as a software licensing preference, draws on two distinct lines of argument. The first is Moglen’s thesis that copyleft functions as a constitutional device for the digital commons: it does not prohibit commercial activity but channels competitive energy away from ownership of the shared substrate and toward the quality of what is built on top of it. The effect is structural, not ideological. By ensuring that modifications to the core remain publicly available when distributed, copyleft prevents the gradual privatisation of the commons through incremental extension, the mechanism that Weber identified as the primary threat to open infrastructure (Moglen, E., “Anarchism Triumphant: Free Software and the Death of Copyright”, First Monday, Vol. 4, No. 8, 1999; Weber, S., The Success of Open Source, Harvard University Press 2004, pp. 83 ff., analysing the political economy of open-source governance and the institutional conditions under which commons can resist enclosure). The second line is the European Union’s own institutional endorsement of copyleft for public sector infrastructure through the European Union Public Licence (EUPL), the only copyleft licence specifically designed by a public authority for use in government software. The EUPL’s compatibility framework and its explicit interoperability provisions reflect the same design logic that the OLRF adopts: an open core that cannot be enclosed, with commercial competition directed toward implementation, service, and value-added capability rather than toward control of the normative substrate. See: European Commission, “European Union Public Licence v. 1.2”, EUPL, 2017; Schmitz, P.-E., “The European Union Public Licence (EUPL)”, International Free and Open Source Software Law Review, Vol. 5, No. 2, 2013, pp. 121 ff. ↩
The multi-stakeholder governance requirement reflects the principle that the evolution of standards that function as public infrastructure must not be captured by any single category of participant. For the governance theory: Ostrom, E., Governing the Commons: The Evolution of Institutions for Collective Action, Cambridge University Press 1990, pp. 90 ff. (design principles for long-enduring commons institutions). For the application to digital commons: Frischmann, B. M., Infrastructure: The Social Value of Shared Resources, Oxford University Press 2012, pp. 255 ff. ↩
The distinction between control over the normative substrate and competition on implementation and service is analogous to the distinction in telecommunications regulation between control over the physical infrastructure (which must remain subject to public governance) and competition in services delivered over that infrastructure (which benefits from market dynamics). See: Regulation (EU) 2018/1972 (European Electronic Communications Code), recitals 3 ff.; for the application to AI infrastructure: Crémer, J., de Montjoye, Y.-A., and Schweitzer, H., “Competition Policy for the Digital Era”, European Commission 2019, pp. 60 ff. ↩
The proposition that the model assignment carries sovereignty implications (not merely technical or constitutional ones) follows from the observation that Models B and C create dependencies that are qualitatively different from the dependencies of Model A. Under Model A, the state depends on evaluation engines and Registry infrastructure, both of which can be built on open-source foundations and operated by the state itself. Under Models B and C, the state additionally depends on AI systems whose development requires resources (training data, compute, research capability) that are currently concentrated outside Europe. The sovereignty question is therefore not whether the state can host its own Registry (it can) but whether it can govern the reasoning infrastructure through which its law is interpreted. That question is unanswered at the European level and constitutes, in our assessment, one of the most significant unresolved strategic challenges of the Law as Code agenda. ↩