Sovereignty is a dial, not a gate.
Most UK organisations using AI are somewhere on the spectrum between total vendor dependency and full operational independence. The Sovereign AI Governance framework measures where — across three vectors: Continuity, Exposure, and Control.
Sovereign Capable AI is an AI estate where an organisation has the continuity provisions, data control, and governance accountability to operate independently of any single vendor — architecturally, not just contractually.
Continuity. Exposure. Control.
Every sovereignty gap traces back to one of three dimensions. Each can be measured, each can be closed, and each has a different remediation path.
Can you keep operating?
Continuity measures whether your organisation can continue to function if your primary AI vendor withdraws access — through commercial decision, regulatory action, or infrastructure failure. Data portability, workflow ownership, and institutional knowledge locked in platform APIs are all continuity risks.
Operations stop or severely degrade if the vendor disappears.
You hold your own data, models, and workflows. The vendor is a preference, not a dependency.
Where does your data go?
Exposure measures what happens to sensitive organisational data when processed by AI — whether it crosses borders, enters third-party training pipelines, or flows through sub-processor chains you cannot fully map. Contractual controls reduce exposure; architectural controls eliminate it.
Data enters third-party infrastructure you do not control or fully understand.
Processing happens within an environment you own or control. Data never leaves your boundary.
Who governs AI decisions?
Control measures whether your organisation — rather than a vendor's platform — governs how AI systems act on its behalf. This requires audit trails, documented escalation paths, human-in-the-loop mandates enforced at the architecture level, and board-level accountability.
The vendor's platform governs AI behaviour. Your visibility into AI decisions is limited.
Documented, tested, board-accountable governance with a complete audit trail. Every action is traceable.
Four bands. One dial.
The three vectors combine into a composite score (0–6) that positions your AI estate on the sovereignty spectrum. Most UK organisations in regulated sectors should target Sovereign Capable.
Operational continuity, data residency, and AI governance all depend on third-party continuity. A vendor change, regulatory action, or contract dispute could materially disrupt operations.
Partial controls exist, but critical dependencies remain. Data may be under partial governance, but vendor lock-in, incomplete audit trails, or untested continuity plans leave meaningful exposure.
The foundations of sovereign AI operation are in place: continuity provisions, data control, and governance accountability documented and tested. The architecture supports independence.
Full operational continuity, data processed within a wholly owned environment, board-accountable governance with complete audit trail. Architecturally independent and defensibly documented.
The Compulsion Test scores you against all three vectors in two minutes.
Take the Compulsion Test →Three converging pressures.
AI sovereignty has become urgent because three independent forces are making vendor dependency materially risky in ways that did not apply three years ago.
Organisations that address the three vectors now build durable capability. Those that do not are making a governance choice — usually without realising it.
EU AI Act
Compliance obligations are now tied to how AI operates — the infrastructure, the governance, the audit trail — not just what it outputs. Vendor-dependent operations may not be sufficient to meet Article 9 obligations.
Vendor-state risk
Geopolitical volatility has made the nationality of AI infrastructure a material consideration for defence, critical national infrastructure, and regulated financial services for the first time.
AI as infrastructure
AI has moved from a productivity layer to operational infrastructure. When AI failure means operational failure, the standards for dependency management must match those applied to any other critical system.
Questions about the framework.
Is residency the same as sovereignty?
No. Data residency — keeping data within a geographic boundary — is a necessary but not sufficient condition for sovereignty. A UK-hosted SaaS system that you do not control, cannot exit, and cannot govern independently is not sovereign. Residency addresses the Exposure vector. Sovereignty requires Continuity and Control as well.
What does 'Sovereign Capable' mean exactly?
Sovereign Capable is the assessed threshold at which an organisation has the architecture, governance, and continuity provisions to operate its AI estate independently. It does not mean operating in complete isolation — it means the capability for independence exists and has been verified. Most UK organisations in regulated sectors should target Sovereign Capable as their operating posture.
What is the difference between AI safety and AI sovereignty?
AI safety concerns the behaviour of AI models — whether they produce harmful, biased, or unreliable outputs. AI sovereignty concerns who controls the operational environment in which AI runs — the infrastructure, the data, and the governance framework. Both matter. Sovereign AI can still be unsafe if model governance is weak; safe AI can still be non-sovereign if it runs on infrastructure you do not control.
Why does sovereignty matter now?
Three converging pressures: the EU AI Act introduces compliance obligations tied to how AI operates, not just what it produces; geopolitical instability makes vendor-state risk material for the first time; and AI has moved from a productivity layer to operational infrastructure — meaning dependency is no longer acceptable at the same level it was when AI was a search improvement.
Know where you stand. Two minutes.
The Compulsion Test scores your organisation against the three vectors and positions you on the spectrum. Take it, see your band, and understand what the gap to Sovereign Capable actually requires.