Prototype concept · Digital Leaders Network Working Draft

The UK AI Sovereignty Index

A measure of how far the United Kingdom retains control over its own artificial intelligence future. Not self-sufficiency, but strategic autonomy: the ability to set terms, switch, audit and exit.

1
On track (Green)
4
Mixed (Amber)
1
At risk (Red)
DIMENSION 01

Compute Control

Amber · Mixed

Can the UK access and operate compute it genuinely controls, rather than capacity it merely rents?

Sovereign capability is emerging through the AI Research Resource at Isambard and DAWN and the new Sovereign AI Unit, and the AI Opportunities Action Plan commits to a large expansion of sovereign compute by 2030. The dependence the other way remains substantial, with most workloads still running on hyperscaler infrastructure exposed to extraterritorial reach such as the US CLOUD Act. Capacity is rising; control is not yet assured.

Lineage: Oxford Insights GARI (AI Infrastructure) · reframed as control, not capacity
DIMENSION 02

Procurement & Lock-in

Red · At risk

Do public bodies buy in ways that preserve the option to switch, or that quietly foreclose it?

Government spends an estimated fourteen billion pounds a year on digital services, much of it concentrated among a small number of foreign suppliers. Exit provisions, data portability and interoperability are too rarely written into contracts, which is the mechanism by which silent lock-in takes hold. This is the dimension where stated ambition and procurement practice are furthest apart.

Lineage: smart-buyer model · exit-by-design (Making AI Work for Britain)
DIMENSION 03

Model Access

Amber · Mixed

On what terms can the UK reach the frontier models its public and private sectors depend on?

Access to frontier capability is broad, but ownership of it is thin: the strongest domestic lab sits within a US-headquartered parent, and dependence on a handful of foreign frontier providers is the norm. The question for sovereignty is less whether the UK can use these models and more whether it can shape the terms on which it does.

Lineage: Tortoise Global AI Index (development) · reframed as terms of access
DIMENSION 04

National Data Assets

Amber · Mixed

Does the UK control and curate the public-sector data that underpins useful, auditable AI?

The National Data Library is a clear statement of intent, and the underlying public-sector data holdings are a genuine national asset. Fragmentation, inconsistent quality and uneven access arrangements mean the asset is not yet usable at the scale the ambition implies. Direction is right; readiness lags.

Lineage: Oxford Insights GARI (Data & Infrastructure)
DIMENSION 05

Skills & Talent

Amber · Mixed

Can the UK develop and, crucially, retain the people on whom autonomous capability depends?

The research base in London, Cambridge, Oxford and Bristol is a relative strength and a real source of advantage. Retention is the soft spot: the pull of better-resourced labs abroad and the acquisition of promising teams together erode the domestic base over time. Supply is healthy; the leak is the concern.

Lineage: Stanford AI Index · Oxford Insights GARI (human capital)
DIMENSION 06

Rule-setting & Audit

Green · On track

Can the UK set the rules for AI in its jurisdiction and inspect the systems it relies on?

This is the UK's clearest area of sovereign strength. The AI Security Institute gives the country a credible, internationally respected capacity to evaluate frontier systems, and the UK's regulatory approach lets it set terms rather than simply inherit them. The standing challenge is converting evaluation capability into enforceable leverage over suppliers.

Lineage: CAIDP democratic values · OECD.AI (governance)

The United Kingdom has rarely been busier on artificial intelligence, and rarely less certain whether all that activity is buying it control or simply commitments.

There is a Sovereign AI Unit with around five hundred million pounds behind it, a national compute capability taking shape across the Isambard and DAWN supercomputers, and an AI Opportunities Action Plan promising a roughly twentyfold expansion of sovereign compute by 2030. Yet the House of Commons Library noted recently that the government still has no overarching policy on digital sovereignty, and that a parliamentary committee found the official response to that absence disappointing. The gap is not ambition. What is missing is a disciplined way to tell whether the ambition is landing. This index is an attempt to provide one.

What this index measures, and what it does not

The index measures a single thing: the degree to which the UK retains control over its own AI future. That is a deliberate choice of definition, and the most important decision in the whole instrument. Sovereignty here does not mean self-sufficiency. A country that built every layer of the stack at home but could not switch supplier, inspect a model, or walk away from a contract would not be sovereign in any sense that matters to a citizen or a minister.

Properly understood, sovereignty is strategic autonomy. It is the capacity to set terms, to switch, to audit and to exit. This framing is consistent with the argument running through Making AI Work for Britain: the route to national advantage is to consolidate demand and diversify supply, not to retreat into autarky. The risk the index is built to detect is therefore not foreign technology as such, but silent lock-in, the slow loss of the option to choose differently later.

Why a new instrument is needed

A mature field of national AI measurement already exists, and the index borrows from it rather than ignoring it. The Oxford Insights Government AI Readiness Index assesses how prepared governments are to use AI. The Tortoise Global AI Index ranks national AI capacity and implementation. The Stanford AI Index tracks research and deployment, the IMF AI Preparedness Index evaluates economic readiness, and the OECD maintains its own comparative index work. Each is credible, and each measures readiness or capability.

None of them measures sovereignty in the sense defined above. The search for a dominant, authoritative AI sovereignty index returns early-stage academic work rather than an established instrument: a recent framework for measuring digital sovereignty presented by Rodríguez Pita, Pérez Martínez and Urueña López at the European Regional ITS Conference, and a small number of calls for exactly this kind of measure now appearing. The field is forming rather than settled. That is precisely the case for Digital Leaders to act: not to clone an existing index, but to fill a genuine gap with a properly founded one.

How the index is built: three foundations

Credibility rests on three borrowed foundations rather than on invention. For method, the index follows the recognised standard for composite indicators set out in the OECD and Joint Research Centre handbook on constructing them. That discipline defines the conceptual framework first, then selects and quality-checks indicators, normalises them, decides weighting and aggregation, and stress-tests the result with sensitivity analysis. For classification, the dimensions descend from the established readiness indices, chiefly the Oxford Insights pillars, rather than being conjured from scratch. For the sovereignty concept, the index draws on the emerging measurement literature and on the UK's own policy record, from the AI Opportunities Action Plan and Compute Roadmap to the parliamentary scrutiny of sovereign capability.

The six dimensions

The instrument resolves sovereignty into six dimensions. Each is drawn from a recognised lineage and then re-cut through the lens of control rather than capacity.

  1. Compute control. Whether the UK can access and operate compute it genuinely controls, shielded from extraterritorial reach, rather than capacity it merely rents.
  2. Procurement and lock-in. Whether public bodies buy in ways that preserve the option to switch, including exit, portability and interoperability provisions.
  3. Model access. The terms on which the UK can reach the frontier models its sectors depend on, not merely whether it can reach them.
  4. National data assets. Whether the UK controls and curates the public-sector data that underpins useful and auditable AI.
  5. Skills and talent. Whether the UK can develop and, more demandingly, retain the people on whom autonomous capability depends.
  6. Rule-setting and audit. Whether the UK can set the rules for AI in its jurisdiction and inspect the systems it relies on, anchored in the AI Security Institute.

How scoring works

Each dimension carries a red, amber or green rating. The colour is only the communication layer. Beneath it sits a small set of named indicators and a documented evidence trail, scored against defined criteria and reviewed by a panel rather than asserted. This is what separates an instrument from an opinion dressed as a scorecard, and it is the part that the OECD and JRC method is there to protect. Few, defensible, repeatable indicators age far better than a sprawling composite.

The natural rhythm is a published baseline followed by a regular review, tied to the conference cycle so that the index becomes a recurring flagship rather than a single slide. A baseline established ahead of the autumn conference could be unveiled there each year, with the movement since the last edition as the story.

What this prototype is, and is not

This is a concept built for discussion, not a finished assessment. The ratings shown are illustrative, prepared to demonstrate how the instrument reads; they are not a verified national judgement. Several of the supporting figures are drawn from secondary summaries and should be confirmed against primary sources before publication. The sovereignty-measurement literature is moving quickly, so the field should be re-scanned before any launch. And the instrument is designed to be owned and hosted by Digital Leaders, with scoring sitting in its research function or an expert panel, not as a personal project of any individual contributor.

Two questions are worth putting to the board directly. First, is control the right organising idea, or would they prefer a broader readiness framing that the existing indices already serve well? Second, where should the line sit between an instrument the network can sustain each year and an ambition it cannot resource? The value of the index will rest less on its first set of ratings than on the discipline and continuity behind them.

Sources and references — to verify against primaries

OECD and Joint Research Centre, Handbook on Constructing Composite Indicators (methodology and user guide). The recognised standard for composite-indicator construction; confirm current edition.

Oxford Insights, Government AI Readiness Index (latest edition). Pillar structure and indicator counts to be confirmed against the primary report.

Tortoise Media, Global AI Index; Stanford HAI, AI Index; IMF, AI Preparedness Index; OECD.AI index work. Capability and readiness measures referenced for classification lineage.

Rodríguez Pita, Pérez Martínez and Urueña López (2025), "From concept to method: a framework for measuring digital sovereignty," European Regional ITS Conference, Edinburgh. Full text to be reviewed.

UK policy anchors: AI Opportunities Action Plan, Compute Roadmap, Sovereign AI Unit, AI Research Resource (Isambard, DAWN), AI Security Institute; House of Commons Library briefing on digital sovereignty. Figures including the Sovereign AI Unit funding and government digital spend to be confirmed against DSIT and parliamentary sources.