Data sovereignty has a decision problem

On June 18, 2025, a French Senate committee put a direct question to Anton Carniaux, director of public and legal affairs at Microsoft France. Under oath, could he guarantee that data on French citizens would never be handed to US authorities without the French government's approval?

"No, I cannot guarantee that," he said. He added that it had never happened.

The answer was candid, and it was not really about Microsoft. The US CLOUD Act of 2018 lets American authorities compel US providers to produce data they control, wherever that data is stored. In our reading, any provider under US jurisdiction would have had to say the same thing. Carniaux just said it out loud, in a room built for the purpose.

More than a year later, buyers are acting on it. Gartner forecast in February that worldwide spending on sovereign cloud infrastructure would reach $80 billion in 2026, up 35.6 percent from 2025. In January, AWS opened its European Sovereign Cloud in Brandenburg, Germany. It is run by EU residents, kept physically and logically separate from the company's other regions, and backed by a planned investment of more than €7.8 billion.

That money buys local facilities, staff and legal entities. It does not buy, on its own, control over what happens to data once software starts making decisions with it.

Why sovereignty is hard to get right

Three things make data sovereignty difficult, and only one of them is about where the servers are.

The first is that the legal ground keeps moving. The EU Court of Justice struck down Safe Harbor in 2015 and its successor, Privacy Shield, in July 2020. The current arrangement, the EU-US Data Privacy Framework, was adopted in 2023. The EU General Court upheld it in September 2025, and the challenger appealed to the Court of Justice that October. Companies that built on the first two rebuilt each time. They may have to again.

The second is that location and jurisdiction are not the same thing. Data can sit in Frankfurt and still be reachable through the company that operates the system. That is the gap the Senate hearing exposed, and it is why European rules increasingly look past residency. The EU Data Act, whose main provisions began applying on September 12, 2025, requires cloud and data processing providers to take measures against unlawful third-country government access to non-personal data held in the EU.

The third gets the least attention. Most sovereignty programs are built around storage: where data may rest. But data rarely leaves a jurisdiction because someone moved a database. It leaves because a process decided to send it somewhere: an integration, a support workflow, a call to an outside service. Those decisions happen well above the storage layer, and residency controls do not see them.

Agents multiply the decisions

AI agents turn that third problem from an edge case into the normal case.

Take a hypothetical. A claims agent at a European insurer runs in an EU region. It reads a customer file, decides it needs a summary and calls a model endpoint. It hands part of the job to a second agent, which calls a translation tool. Each step is reasonable. Each one is also a decision about where customer data goes next, made at runtime, by software, with nobody approving it in the moment.

The infrastructure is sovereign. The workflow may not be. If the model endpoint or the translation tool sits outside the EU, the data has crossed a border, and the residency controls had nothing to say, because the storage never moved.

Access control does not close this gap either. It decides whether an agent may reach a system. It does not decide whether what the agent is about to do with that system is something it was ever allowed to do.

Where authority governance fits

Authority governance treats jurisdiction as part of an agent's mandate, not only as a property of the infrastructure underneath it.

We define authority in three parts. Decision rights: what an agent may decide on its own. Constraints: the limits on those decisions. Accountability: who answers for the outcome, and the record that shows what happened.

Decision rights can state which data an agent may send outside its home jurisdiction, if any. Constraints make the rule specific: data classified as EU-resident may go only to endpoints in approved regions, and any request to send it elsewhere stops and waits for a person. When the claims agent tries to call an endpoint outside the EU, the check happens on that action, at that moment. The outcome is permit, escalate or deny, and the attempt is recorded either way.

That record is, in our view, the piece most sovereignty programs are missing. When a regulator or a customer asks where a piece of data went and who authorized it, a storage map answers half the question. A decision record answers the other half: which agent acted, what it was allowed to do, what it tried and what was decided.

There are limits, and they matter. Authority governance does not change which country's law applies to a provider, and it cannot block a lawful order served on that provider. Sovereign infrastructure, contracts and customer-held encryption keys are the tools for that. What authority governance adds is control over the movement those measures were not designed to see: the decisions software makes on its own, many times a day, with no one watching.

Countries and companies are spending heavily to decide where their data lives. The question agents make urgent is who decides where it goes next.

---

Sources. The Register, July 25, 2025, on Microsoft France's testimony to the French Senate on June 18, 2025. Gartner, "Gartner Says Worldwide Sovereign Cloud IaaS Spending Will Total $80 Billion in 2026," February 9, 2026. Amazon, AWS European Sovereign Cloud launch announcement, January 2026. WilmerHale, December 1, 2025, on the General Court's Latombe ruling and the appeal to the Court of Justice. Hunton Andrews Kurth on the EU Data Act's application date and Article 32. Regulatory statements are our reading, not legal advice.