The hidden politics of AI: sovereignty in a platform world

AI sovereignty is no longer an abstract policy concern - it is now a live operational risk. When critical institutions can be constrained by platform providers, and AI systems embed governance choices by design, the question shifts from “who regulates AI?” to “who controls the systems we depend on?” The signals have been visible for years. Now they are unavoidable.

ICC - the canary in the coalmine

When Microsoft reportedly blocked email access for International Criminal Court officials, the reaction in many policy and technology circles was surprise. And it shouldn’t have been.

The real question is not “how could this happen?” but “why didn’t you read the tea leaves?” Because the signals have been visible for years. We have been steadily outsourcing critical institutional capability - communications, data, infrastructure - to a handful of private technology providers operating under specific national jurisdictions.

This was always going to collide with sovereignty.

The illusion of neutral infrastructure

For a long time, cloud platforms and software providers were treated as neutral infrastructure - like electricity or plumbing. Reliable, scalable, and politically inert. And that assumption no longer holds.

Technology platforms are subject to the legal, political, and strategic priorities of the jurisdictions in which they are headquartered. When push comes to shove, they cannot be neutral. They will comply with state power.

The ICC incident is not an anomaly. It is a case study.

If an international legal body can have its access constrained by a private company, then sovereignty is no longer just about territory or law. It is about who controls the systems your institutions depend on.

AI changes the stakes

Now layer AI on top of this. We are not just talking about email or cloud storage anymore. We are talking about systems that mediate decision-making, shape information flows, and increasingly act on behalf of users and organisations.

The recent “Anthropic directive” controversy - where system-level constraints and behavioural shaping became visible - should be read in this context. It is not just a product design issue. It is a governance signal.

• Who decides what an AI system can say, prioritise, or refuse to do?
• Who sets those boundaries - and under whose authority?

These are not abstract questions. They are questions of control.

Sovereignty is becoming computational

Sovereignty is no longer just exercised through borders, laws, and institutions. It is increasingly exercised through code, models, and infrastructure.

This creates a new layer of dependency:

• Dependence on foreign-owned compute infrastructure
• Dependence on proprietary foundation models
• Dependence on platform-level policy decisions embedded in AI systems

Each of these dependencies introduces potential points of control - or failure.

And unlike traditional supply chains, these dependencies are often opaque. The governance mechanisms are buried in terms of service, API constraints, and model behaviour that is difficult to audit.

Nations, culture, and strategic advantage

What is often missing from the AI sovereignty conversation is the role of national character - what we might think of as the “folk” layer of strategy.

Countries do not approach AI in a vacuum. They bring with them distinct traditions about authority, knowledge, risk, and control. These shape not only regulation, but how AI systems are designed, deployed, and constrained.

Some states approach AI as an instrument of state coordination and long-term strategic advantage. Others treat it as a market-driven capability, tempered by rights-based safeguards. Still others are attempting to reconcile innovation with deeply embedded social or cultural norms.

This matters because AI systems carry these assumptions with them.

The idea of “respect” for AI - how much autonomy it is given, how tightly it is controlled, how much it is trusted in decision-making - varies across jurisdictions. In some contexts, AI is treated as a tool to be tightly bounded. In others, it is positioned as a collaborator or even a strategic actor.

These differences are not philosophical curiosities. They translate directly into national advantage.

• Countries that align AI development with industrial policy can accelerate capability at scale
• Countries that embed AI deeply into statecraft gain leverage in intelligence, defence, and diplomacy
• Countries that fail to develop internal capability risk becoming rule-takers in systems they do not control

In this sense, AI sovereignty is not just defensive. It is also about ambition.

The strategic blind spot

Many organisations - and governments - are still treating AI adoption as a purely technical or economic decision. It is not.

It is a sovereignty decision. And decisions made now will determine national fates for decades.

Choosing a model provider, a cloud platform, or an AI development stack is also choosing a set of embedded governance assumptions. It is accepting constraints that may only become visible under stress.

The ICC and Anthropic examples illustrate what happens when those constraints surface unexpectedly. The Anthropic directive illustrates that they are already being designed into systems from the outset.

What “seeing” looks like

So what does it mean to actually “see” what is happening? It means recognising that:

• AI systems are not neutral tools; they are governed artefacts
• Platform providers are not just vendors; they are geopolitical actors
• Technical architecture decisions are also political decisions

And it means acting accordingly.

This does not necessarily imply full technological autarky - that is neither realistic nor desirable. But it does require a more deliberate approach to dependency management, capability development, and governance design.

At a minimum, this includes:

• Understanding where critical dependencies sit in your AI stack
• Assessing jurisdictional exposure and legal risk
• Building optionality into systems and vendor relationships
• Investing in internal capability to evaluate and govern AI systems

The end of naivety

I have, and will continue to argue, that we are moving out of a period of technological naivety. The idea that we could build globally integrated digital systems without confronting questions of power and control was always optimistic. AI is simply forcing the issue.

The tea leaves were always there: in export controls, in data localisation laws, in platform moderation policies, in the gradual securitisation of technology supply chains. The ICC incident simply made it visible.

The behaviour of AI systems is making it unavoidable. The question now is not whether sovereignty matters in AI. It is whether institutions - and nations - are willing to act as if it does.