UK AI Strategy

When asked about Britain's AI strategy, many default to the EU's regulatory framework - this is misleading.

The UK needs an AI strategy organised around three sequential priorities: sovereignty first, then facilitation, then governance. This ordering is not arbitrary. You cannot sensibly govern what you do not control, and you cannot facilitate what you have not chosen to build. Critics who respond to this argument by pointing to the EU AI Act as a ready-made solution have, perhaps unintentionally, started at the end and worked backwards.

Governance without sovereignty

The AI Act is a risk-classification and compliance framework. It categorises AI systems by the potential harm they pose and attaches corresponding obligations. What it does not do — and was never designed to do — is answer the prior question of whether a country actually controls the AI infrastructure it depends on. The UK's own Compute Roadmap has warned that without urgent action, Britain risks being "over-reliant on foreign infrastructure." No amount of regulatory labelling changes that material reality. Sovereignty requires capability, not just compliance categories.

Crucially, the Act exempts military purposes and national security applications from its scope entirely. For a government treating AI as a strategic and security asset, a framework that carves out the most sensitive domain by design is incomplete from the outset.

"The Act says nothing about compute ownership, semiconductor supply chains, or the geopolitical conditions under which access to frontier AI could simply be withdrawn."

It does not build; it only constrains

The facilitation pillar of a serious AI strategy requires affirmative choices: investment in sovereign compute, energy infrastructure, planning reform, and industrial incentives to anchor AI companies at home. The Act contains none of this. Academic analysis has noted that the Act risks being perceived as a barrier to innovation if not balanced with support for research and development — a concern serious enough that the EU has since been forced to propose a separate technology sovereignty package to address the industrial gaps the Act left untouched. The EU itself has implicitly conceded that regulation alone was never sufficient.

A confused concept of sovereignty

Where the Act does gesture at sovereignty, the concept is underspecified. Scholars have identified three unresolved tensions within EU AI sovereignty: whether it pits the bloc against other major powers or citizens against large tech companies; whether it is meant to boost competitive position or resist that logic altogether; and whether it benefits European citizens primarily or carries a global responsibility. A UK strategy cannot import that ambiguity. It needs a settled answer to the sovereignty question before governance instruments can be properly calibrated.

The wrong sequence

Recommending the EU AI Act as a response to an argument about sovereignty and facilitation is a little like being asked how to build a house and answering with a description of the planning regulations. The regulations matter — but they are downstream of the decision to build, the choice of site, and the securing of materials. Governance is the final layer, not the foundation. Britain's AI strategy needs to get the order right.