Summary

• The Council for Science and Technology recommends a national AI Sovereignty Plan identifying which technologies the UK should develop, share, or access through international suppliers.
• It proposes a single accountable government lead, public interest pilots, stronger skills and research capabilities, and better access to strategically important datasets.
• The advice was submitted on 30 July and published on 19 October 2026. The four recommendations are not an adopted government programme.

Britain should decide which artificial intelligence capabilities it needs to control directly and where it can safely depend on international suppliers, according to government science advisers calling for a national AI sovereignty strategy.

The Council for Science and Technology recommends mapping the UK’s existing computing infrastructure, models, data, skills, and institutional capabilities before determining which investments would strengthen the country’s ability to develop and use AI independently.

The advice, submitted to minister Kanishka Narayan on 30 July and published on 19 October, argues that domestic capability cannot be measured solely by whether advanced AI models are developed within Britain. The council wants government to consider the wider systems and expertise required to operate technology reliably, capture economic benefits, and respond when access to external services is disrupted.

The Council for Science and Technology makes four recommendations covering a national plan, government accountability, investment in domestic capabilities, and access to international models. They are advisory proposals rather than decisions already adopted or funded by ministers.

Sovereignty requires choices across the technology stack

The council defines AI sovereignty in terms of the UK’s ability to make meaningful choices about how AI is developed, accessed, and used while preserving resilience during adverse circumstances.

Under that definition, ownership of a domestically developed model is only one possible component. Computing facilities, access to suitable data, technical expertise, organisations capable of operating AI systems, and the governance arrangements surrounding deployment all contribute to national capability.

A British model, for example, might still depend on specialised chips, cloud services, or software maintained overseas. Conversely, an internationally developed model could potentially be operated under arrangements that provide adequate control for a particular domestic application.

The council therefore recommends an AI Sovereignty Plan based on an assessment of the complete technology ecosystem. Government would identify existing capabilities and dependencies, then decide which elements should be owned domestically, developed collaboratively, or accessed through resilient external arrangements.

Those choices would involve different costs and commercial consequences. Attempting to replicate every advanced capability within Britain could demand substantial investment without guaranteeing comparable technical performance.

Depending entirely on external providers, however, could expose important services to decisions about access, pricing, technical changes, or international availability over which UK organisations have limited influence.

The council is advocating a selective approach that weighs those risks against the benefits of continued participation in the international AI market.

Public interest pilots would test the available options

Rather than treating sovereignty as a purely strategic objective, the council proposes practical experiments involving public interest applications in areas such as national security, healthcare, and education.

Each pilot would begin with the problem to be solved and examine the data, model capabilities, and surrounding systems needed to address it.

The recommendations also require consideration of the value that could be captured and how the results would be measured, encouraging decisions based on operational evidence rather than assumptions about the benefits of building national technology.

Different applications could require different arrangements. A narrowly defined administrative task might be handled using a smaller specialised model, while scientific research could require access to more powerful internationally developed systems.

The council suggests testing such alternatives rather than assuming that the largest general-purpose model offers the best solution for every public service.

Operating those pilots would also create opportunities to develop technical experience within government and its partner organisations, including expertise in model evaluation, data preparation, deployment, and procurement.

The advice calls for public engagement in the design of the proposed pilots, particularly where the technology affects services delivered to citizens or uses information held by public institutions.

That requirement would add a governance component to decisions that might otherwise be dominated by technical performance or procurement costs.

No pilot funding, final participating departments, or implementation timetable is established by the published advice. Those decisions would require a government response and subsequent delivery arrangements.

One accountable lead would coordinate government decisions

The council’s second recommendation concerns the organisation of AI sovereignty policy across Whitehall.

Responsibility for AI involves departments dealing with public services, national security, industry, research, education, and public finances. Their decisions are connected because investments in one part of the technology system can influence requirements elsewhere.

Funding computing infrastructure, for example, may have limited value if organisations cannot obtain appropriate data or lack the specialists required to develop useful applications.

Similarly, separate departments could procure technologies that create incompatible platforms or different dependencies on external suppliers, increasing the difficulty of managing them consistently.

The council recommends appointing a single accountable owner to coordinate those choices, identifying the head of the Cabinet Office AI Taskforce as a possible candidate.

That person would bring departments together and help shape investment and delivery decisions, supported by strategic and technical advice.

The recommendation does not itself appoint anyone or transfer departmental responsibilities. Its practical effect would depend on the authority, resources, and decision-making arrangements established if ministers choose to implement it.

Skills and public data form national infrastructure

The third recommendation calls for investment in capabilities that extend beyond software and computing hardware.

The council wants more opportunities for specialists to develop experience through career pathways and secondments, complementing existing initiatives such as TechFirst.

AI expertise develops through practical work as well as formal education, particularly where teams must evaluate model behaviour, manage complex datasets, and integrate systems into operational environments.

Retaining that experience within the UK could help businesses and public institutions make informed technology decisions without relying entirely on external contractors or platform suppliers.

The council also recommends strengthening evidence about AI adoption and productivity, allowing investment decisions to be assessed against measurable outcomes.

Better evidence would help distinguish between organisations purchasing additional software and those achieving lasting improvements through changes to business or public service processes.

Publicly held information is another component of the proposal. The council wants government to identify important UK datasets and test access arrangements in at least two valuable areas, suggesting health and creative data as possible examples.

Such access would still need to comply with legal obligations concerning privacy, security, confidentiality, and intellectual property. The recommendation does not call for unrestricted release of sensitive public records.

The council also supports continued research into fundamental computing technologies, including lower-power approaches that could reduce the energy requirements of future systems.

Foreign models would remain part of the UK’s approach

The fourth recommendation explicitly recognises that access to leading international AI models remains important.

Rather than attempting to replace those systems completely, the council proposes developing selected domestic capabilities that provide alternatives for particular applications while retaining trusted access to the most advanced global technology.

It suggests using public interest pilots to compare open-weight models, smaller domestically developed systems, and other approaches to obtaining AI capability.

Government would also allocate some national computing capacity to those experiments, allowing researchers and public institutions to examine alternatives without immediately committing to a single technology supplier.

The advice identifies a proposal from the Alan Turing Institute to develop an open-source sovereign base model as an option worthy of consideration.

The institute envisages a model governed within the UK and made available as a public resource for researchers, smaller businesses, public services, and other users. The council does not present that proposal as an approved government project or commit funding to its development.

The approach reflects a distinction between preserving choice and pursuing complete technological independence. Britain could continue using powerful international systems while investing in domestic capabilities where they provide meaningful resilience, operational control, or commercial opportunity.

For businesses, the consequences of any future strategy would depend on how it influences access to computing, public data, research expertise, and technology procurement.

The council’s advice was prepared in July and has now entered the public record. Whether it becomes a national programme will depend on ministerial decisions, the allocation of resources, and the evidence generated by any practical pilots subsequently approved.