The answer will not be the same for every application.
Losing access to an AI tool used for routine administrative work would be disruptive. Losing access to AI embedded within public services, critical infrastructure, defence, financial systems or national security operations could have far more serious consequences.
As AI moves from experimentation into the operational fabric of the country, access is becoming a question of national resilience.
Dependency is not inherently a weakness
The UK cannot and should not attempt to build every component of the AI technology stack domestically.
Britain benefits from international partnerships and access to technology developed elsewhere. UK organisations should continue to use the best models available.
Dependency becomes dangerous when it is poorly understood, excessively concentrated or unsupported by a credible continuity plan.
The issue is not where an AI model was developed. It is whether an essential UK operation could continue if access changed.
A provider could restrict access for particular sectors or uses, withdraw a capability, change its commercial terms or limit where a model can be deployed. Geopolitical relationships may also change more quickly than government procurement or technology strategies.
Access cannot therefore be treated as a permanent guarantee.
Not every capability needs to be sovereign
A serious national strategy must avoid two extremes.
The first is assuming that because the UK can access leading AI models today, that access will always be available on the same terms. The second is believing Britain must recreate every frontier model, cloud platform and supporting technology itself.
Neither is realistic.
Government should instead identify the AI capabilities essential to the continued operation of the country and determine what level of dependency is acceptable in each area.
That assessment should distinguish between applications where disruption would create inconvenience, those where it would damage productivity and those where it could prevent an organisation from performing its core function.
This provides a more practical definition of Sovereign AI. Sovereignty should not mean technological isolation. It should mean retaining sufficient control, capability and choice to continue operating when technology, suppliers or political circumstances change.
Operational resilience is the outcome. Sovereign capability is one of the means by which we achieve it.
Britain must understand its critical dependencies
The UK cannot reduce its exposure until it understands where that exposure exists.
Government should map where frontier AI is already supporting critical national functions and where it is likely to become operationally important over the next five to ten years.
This must extend beyond direct government use. Dependencies will exist throughout supply chains, outsourced services, regulated industries and the technology companies supporting public-sector operations.
The assessment should also examine the wider ecosystems around AI models, including infrastructure, data services, technical expertise and commercial relationships.
The test should be practical: if access changed, could the organisation continue to fulfil its mission?
Once those dependencies have been identified, the UK can determine where domestic capability, strategic partnerships or greater operational control are required.
Britain is unlikely to succeed by attempting to reproduce the scale of every global frontier AI company. It can, however, develop differentiated capabilities in areas where it has operational expertise, valuable data, research strengths and clear national requirements.
The objective should not be to build a British version of every AI product. It should be to ensure the UK possesses the capabilities it needs to remain economically competitive and operationally resilient.
Resilience must shape procurement
Government procurement must support this objective.
AI used in critical environments should be assessed on continuity, control and resilience alongside performance, price and speed of deployment.
Buyers should understand where operational data is held, who governs the system, what happens if access or commercial conditions change and whether the organisation can continue operating during disruption.
The public sector must also avoid allowing short-term adoption targets to create long-term strategic dependencies.
Rapid adoption may deliver immediate productivity gains. But if essential services become reliant on technology they cannot govern or sustain, the UK will have exchanged one legacy problem for a new vulnerability.
From an access review to a resilience strategy
The government is right to examine the consequences of losing access to frontier AI models. But calculating the potential economic impact will not be enough.
The UK needs an AI resilience strategy that:
- Identifies the AI capabilities becoming critical to essential operations.
- Maps dependencies across government, infrastructure and strategically important industries.
- Defines acceptable levels of external dependency.
- Determines where sovereign capability or greater operational control is required.
- Uses investment and procurement to strengthen resilience.
- Tests whether critical organisations could continue operating if access changed.
This is not an argument for turning away from international partners. It is an argument for entering those partnerships from a position of clarity.
The countries best placed to benefit from AI will not necessarily be those that own every model or component. They will be those that understand their dependencies, protect their critical capabilities and can continue operating when conditions change.
The UK now has an opportunity to establish what meaningful AI resilience looks like before dependency becomes disruption.
The question is not simply whether Britain can access the best AI in the world today. It is whether we are building the resilience to keep operating tomorrow.
