As AI agents gain greater autonomy to access data and initiate actions, data sovereignty can no longer depend on permissions applied after collection. Control over data visibility must become enforceable at the point where data is generated.
By Elena Pasquali
Agentic AI Changes the Rules of Data Control
Agentic AI is changing the role of artificial intelligence in digital systems. AI agents can access multiple data sources, interact with applications and other agents, make decisions and initiate actions with limited human intervention.
This creates major opportunities, but it also changes the rules of data control. As AI agents become increasingly autonomous, controlling who can access which data, for which purpose and for how long becomes a foundational architectural requirement rather than a compliance exercise.
From Platform Control to User Sovereignty
In traditional architectures[1], data is generated by an application or connected device and then collected by a cloud platform, data lake or other centralized environment. This architecture delegates policy enforcement to the collecting platform. While the data generator may retain formal rights over its data, the platform ultimately exercises technical control over who can access it, under which conditions, and for how long.
With agentic AI, this limitation becomes significantly more serious. AI agents can continuously access data, combine it with multiple sources, and trigger autonomous actions across interconnected systems. The key question is therefore no longer only where data is stored, but who determines which agent can see which data, for which purpose, and for how long.
True data sovereignty[2] requires shifting policy enforcement from the platform to the point of data generation. Control over data visibility must move from the point of collection to the point of generation – the application, machine or connected device producing the data.
Data should be protected before entering the infrastructure and remain inaccessible by default. Visibility should be granted only to explicitly authorized AI agents and other recipients, and access should be revocable at runtime, immediately preventing any future visibility.
Enforcing visibility control at the origin
This changes the role of the infrastructure. Cloud platforms, brokers and AI environments may continue to transport, store and process data, but they no longer determine who can see it. Control over visibility is enforced cryptographically at the origin.
Identity and access management remain essential, but they are not sufficient. Identity can establish who an agent is, and platform permissions can define which resources it may access. Neither alone guarantees that the data generator retains direct control once the data enters a third-party environment.
Origin-side enforcement closes this gap. Different data streams can be made visible to different agents without exposing them to the distribution infrastructure or to unauthorised recipients. Governance, identity, policy, audit and orchestration may remain centrally coordinated, while enforcement is distributed across the data sources themselves. These functions cannot make data visible unless the corresponding policy is also enforced cryptographically at the origin[3].
From Principle to Practice
Agentic AI will give machines unprecedented autonomy to act on data. That autonomy must be matched by an equally scalable ability to control data visibility continuously and at the point of data generation.
The decisive question is no longer where data resides, but where control is enforced. As long as policy enforcement remains delegated to the collecting platform, user sovereignty remains a legal principle rather than an operational reality. Only when control over data visibility is enforced where data is generated does user sovereignty become an architectural capability.
A Personal Reflection
When I first began working on digital sovereignty, my focus was almost entirely on a governance question: who really controls access to the data generated by our devices? Much of the debate was centered on the jurisdiction of cloud providers and the location of data. Those are important issues, but over time I became convinced that they were not the fundamental problem.
The real question is not where data is stored, but where control is exercised. If the authority to decide who can access and use data remains concentrated in centralized platforms, then neither sovereignty nor security can ever be fully achieved. That conviction ultimately led us to develop the DVCO (Data Visibility Control Overlay[4]), based on a simple architectural principle: move cryptographic control from the point where data is collected to the point where it is generated.
As DVCO evolved, it led me to a broader realization. The same centralized mechanisms that concentrate control over data also concentrate cyber risk. If compromising a single cryptographic control point can expose an entire digital ecosystem, then resilience cannot simply mean defending that point more effectively – it requires eliminating it as a systemic dependency. Once every connected device becomes an autonomous cryptographic trust domain, the compromise of a single component no longer threatens the entire ecosystem.
Today I see digital sovereignty and cyber resilience not as separate objectives, but as two outcomes of the same architectural principle. DVCO was conceived to restore control over data to the citizens, organizations and institutions that generate it; in doing so, it also provides a new foundation for cyber resilience in an increasingly interconnected world.
[1] By traditional architectures, this essay refers to centralized data-sharing models in which data is collected and managed by a central platform that controls storage, access, and policy enforcement.
[2] Data sovereignty refers to the ability of a data owner or provider to retain control over its data throughout its lifecycle, including determining who may access it, for what purpose, under what conditions, and for how long, while ensuring that these usage policies are technically enforceable.
[3] Christian Jung and Jörg Doerr, Data Usage Control, in Designing Data Spaces – The Ecosystem Approach to Competitive Advantage, Springer, 2022
[4] Ecosteer’s core technology. The author is co-founder of the company
