Summary
- A Council for Science and Technology letter written in July was publicly released on 19 October.
- The advisers recommend a national plan, clearer governance, investment in domestic capability, and trusted access to leading models.
- Their definition of AI sovereignty emphasises strategic choice and resilience, not complete technological self-sufficiency.
The UK government’s senior science advisers have recommended a coordinated strategy for preserving national control over artificial intelligence capabilities, warning that decisions about external technological dependencies will shape long-term resilience and public trust.
The advice was prepared by the Council for Science and Technology in a letter dated 30 July and published by the government on 19 October.
Its recommendations call for a national AI Sovereignty Plan, stronger coordination across government, investment in selected domestic capabilities, and continued trusted access to leading international AI systems.
The advisers do not equate sovereignty with developing every component of the AI supply chain inside Britain. Instead, they define it in terms of retaining sufficient agency to pursue national objectives, realise public and economic value, and maintain resilience when external conditions deteriorate.
That approach recognises that technological independence is not an absolute condition, but that dependencies must be understood before they become obstacles to national decisions.
Four recommendations for government
The council’s first recommendation is the creation of a UK AI Sovereignty Plan based on an accurate assessment of the AI ecosystem.
Such an assessment would need to identify capabilities available domestically, areas where international partnerships are necessary, and dependencies that could materially constrain public or commercial activities.
The second recommendation concerns governance and coordination. Different parts of government make decisions affecting computing infrastructure, research, security, procurement, skills, and technology markets. Without a common framework, those decisions may create overlapping investments or leave important dependencies unexamined.
The advisers also recommend investing in the foundations of long-term domestic capability, including technical skills, data, and public trust.
The final recommendation is to maintain access to leading international AI models while developing capabilities where the UK has a credible opportunity to contribute or lead.
This combination avoids treating imported technology as inherently unacceptable. It instead proposes deliberate choices about which capabilities should remain available under adverse circumstances.
Resilience across the AI supply chain
The security dimension extends beyond model behaviour. AI services depend on data centres, compute infrastructure, electricity, networking, software development tools, datasets, and commercial agreements with platform providers.
An organisation may have direct control over its own applications while depending on third-party models, hosting, or specialised computing hardware for those applications to operate.
If a provider changes access conditions, suffers a major outage, or becomes unavailable because of political or commercial restrictions, dependent systems may have limited alternatives.
These risks are not unique to AI, but advanced models introduce additional concentration because a relatively small number of suppliers provide some of the most capable systems.
Substituting a different model may require changes to software integrations, security controls, evaluation procedures, data handling, and operational workflows.
Consequently, possessing access to several model providers does not necessarily establish that an organisation can change between them quickly or safely.
National sovereignty policy must therefore distinguish the ability to procure a service from the ability to maintain essential functions when that service cannot be used.
Strategic control rather than isolation
For public authorities and regulated businesses, external AI dependency raises questions about information governance, service continuity, contractual controls, and the ability to audit critical decisions.
Some systems may process sensitive information or support functions that require assurance beyond the general security promises of a commercial supplier.
Other uses may be less sensitive and better served by internationally supplied technology, particularly where domestic alternatives would be costly or technically inferior.
The council’s recommendations suggest that government should make those distinctions systematically instead of seeking full national ownership of every model, chip, or data centre.
The advice also places public trust alongside technical capability. Domestic infrastructure does not automatically establish trustworthy AI behaviour, while reliance on foreign providers does not automatically make a service insecure.
Assurance depends on how systems are governed, evaluated, operated, and recovered, as well as where their components originate.
The council’s letter offers strategic recommendations rather than announcing a funded government programme or a final policy decision. Ministers have yet to set out how the proposals will be translated into specific investments and procurement expectations.
Publication of the advice establishes a framework for those decisions, with resilience under duress identified as an explicit consideration in the UK’s approach to AI sovereignty.
