As AI agents move data across systems, the rules must move with it

HISTORICALLY, data sovereignty has been synonymous with geography, focused on location-based questions. Where is data stored? Where is it permitted to go? Who has access to it? What regulatory bodies govern its use?

Those questions remain critical, but agentic AI has created sovereignty issues on its own.

Agents do more than read data. They call tools, switch between models, change course and act on their own. That pushes sovereignty past who can reach data and into what a system does with it. 

Organisations need to set rules for whether an AI system can analyse a dataset, combine it with other sources, hand it to another model or move it to another environment. The orchestration layer that directs agents is where those rules get enforced.

Sovereignty has become a question of authority, not just location. 

In Lenovo’s work with enterprise customers, there are fewer questions about where data sits and more about what autonomous systems may do with it. That shift changes how companies must govern sovereignty.

Why AI agents need identity and access controls

Identity and access management systems were built to govern people. Enterprises now run human and non-human identities, such as service accounts and bots, side by side, and both handle sensitive information. Agentic AI makes this harder because agents can act on their own.

An agent should never have more authority over information than the person it works for. For regulators and risk teams, that means managing agents under the same rules as people, with permissions that mirror existing governance.

Without that, an employee could be barred from moving data into a particular environment, only for an agent to find a technical route that makes the same move. To the agent, that is problem-solving. To the enterprise, it is a sovereignty failure.

Consequently, organisations must treat agents as both endpoints and actors, making identity and access management foundational to sovereign AI.

This also reinforces why identity management must evolve beyond authenticating users. Organisations increasingly need consistent identity, policy and lifecycle management for both human and non-human actors.

Setting boundaries on how AI agents use data

Identity management alone is insufficient. The same adaptability that makes agentic AI valuable also makes it a sovereignty challenge.

Organisations cannot just tell an agent what information it can read. They must also explicitly define these:

• What it cannot access
• What it may and may not do with permitted information
• Where that information can travel
• Which environments are off-limits 

These negative boundaries are critical to ensuring an agent does not act autonomously to take an otherwise prohibited action to accomplish its objective. 

Agents need autonomy to solve problems developers have not anticipated, which is core to their value. But that autonomy must have bounds, with enough room to create value and explicit limits on data and actions.

That makes sovereignty an infrastructure requirement as much as a compliance one. 

Identity controls must govern human and non-human actors; network boundaries must prevent prohibited data movement; and data classification must account for information that may be categorised differently across national regulatory systems. Data governance and sovereign AI have become inseparable.

For many organisations, these controls will ultimately become infrastructure capabilities rather than policy documents.

Maintaining AI sovereignty across cloud and local infrastructure

The sovereignty challenge is growing as AI infrastructure becomes increasingly distributed. Organisations must consider where data is processed, not just where it is stored.

AI workloads often span local infrastructure, data centres, cloud resources and hybrid environments. A workload could originate in one jurisdiction, interact with a model in another environment, invoke additional processes and return information to its starting point.

Organisations must know the path their data takes along the way.

As a result, they need consistent sovereignty controls across infrastructure and geographies. The same dataset cannot be governed by incompatible rules as it moves between environments.

The emerging model is one in which sovereignty controls persist as data, while workloads and models move through distributed infrastructure. The rules need to follow the data.

This is one reason hybrid infrastructure is becoming increasingly important. Organisations need sovereignty controls that remain consistent regardless of whether workloads execute at the edge, on-premises or in the cloud.

Why AI sovereignty requires more than regulatory compliance

This evolution is happening faster than the regulation tasked with governing it.

Most existing rules contemplate what people and organisations can do with information, and agentic AI introduces systems capable of autonomously performing those actions. 

Meanwhile, multinational businesses must reconcile overlapping regional, national, industry, contractual and customer requirements.

Regulators will continue establishing parameters for data storage, processing and movement. However, businesses cannot wait for these policymakers to provide detailed guidance for agentic use cases. The technology simply moves too fast.

As a result, regulatory compliance should be viewed as a floor, not a substitute for a sovereign AI strategy.

Furthermore, sovereignty cannot belong to any department within an enterprise. It must intersect with legal, cybersecurity, data governance, infrastructure and AI teams. Leaders must ensure governance, security, technical and risk-management functions all translate sovereignty requirements into practical controls.

The first era of digital sovereignty was primarily about location: where information lived, who could access it and which rules apply?

The agentic era adds new questions. What can an autonomous system do with that information, and where can it take it?

Agents need autonomy to be effective, so organisations must ensure that autonomy operates within specific, pre-set guardrails.

While the practice has become increasingly complex, the principle behind AI sovereignty in the agentic era is straightforward. Agents must be treated like their human peers, and the data they handle must have strict governance rules that follow it wherever it goes. 

Lenovo helps enterprises design, deploy and govern AI agents across client, edge and cloud. Explore Lenovo AI Services online.

This article is written by Christopher Campbell, director of AI governance, Lenovo.