The UK government's senior science advisers want ministers to draw up a UK AI sovereignty plan that decides, technology by technology, what Britain should build itself and what it can safely buy from abroad. The advice comes in a letter from the Council for Science and Technology (CST) to Kanishka Narayan, the minister for artificial intelligence, published on GOV.UK on 19 October 2026.
The letter is dated 30 July 2026, a little over a week after Narayan became the first AI minister to attend cabinet. It was copied to the Prime Minister, the Chancellor of the Exchequer, the Cabinet Secretary, the chair of the Prime Minister's AI Taskforce and the secretaries of state for business, defence and culture.
Its central warning is blunt. Without urgent action, the co-chairs write, the UK risks becoming "a high-value AI customer" rather than a country that shapes the technology and how it is used.
What the UK AI sovereignty plan would cover
The CST is the government's top-level independent advisory body on cross-cutting science and technology questions. Its co-chairs, Dame Angela McLean and Lord Browne of Madingley, signed the letter. Nine council members are credited with the work, including Professor Sir Nigel Shadbolt, Professor Nick McKeown and Priya Lakhani.
The council defines AI sovereignty as having "sufficient agency across the AI value chain" to deliver the outcomes the UK chooses, capture economic and public value, and stay resilient under pressure. It does not treat sovereignty as owning a national chatbot. The letter calls it "a whole-system challenge" that cannot be met by funding isolated technologies or projects.
The first recommendation is a rapid map of the UK's AI ecosystem. That map would cover computing power, data, skills, models, governance and public adoption, and would record what already exists, where the UK depends on others and where the gaps are. The council wants the map used as a decision tool, sorting capabilities into three groups:
- capabilities the UK should own outright;
- capabilities it should build with partners;
- capabilities it can access from abroad, provided that access is resilient.
Pilots in security, health and education
Alongside the map, the CST proposes fast pilots in public-interest areas, naming national security, health and education as examples. Each pilot would come with a plan for public engagement so that people have a say in the design. The letter sets six questions each pilot should answer:
- What problem is being solved?
- What sovereign data does the UK hold for it?
- What model capability is needed?
- What system has to be built around the model?
- What value can the UK capture?
- How will success be measured?
The questions push the debate away from model size and towards data and delivery. In the council's words, data is valuable only when it is managed, secured and used well.
One owner inside government
The second recommendation deals with a familiar Whitehall problem: AI work spread across departments with nobody in charge. The CST asks the government to name a single accountable owner, suggesting the head of the Cabinet Office AI Taskforce, to pull departments together and shape funding and delivery. It also wants strategic and technical advice combined, so that pilots get informed judgements on which models, datasets and computing options to use.
This lands at a moment of upheaval in how the UK runs technology policy. According to reporting by The Next Web, the Department for Science, Innovation and Technology was folded into a new Department for Business, Innovation, Science and Trade in July, and the limits of Narayan's own remit had not been fully set out when he took the job.
Skills, data and compute
The third recommendation is about long-term capability rather than quick wins. The council argues that skills and commercial value build up where AI systems are actually built and used, so the UK needs career paths that keep experts in the country. Its proposals include:
- an AI skills and secondment route that complements the existing TechFirst programme;
- a stronger evidence base on AI adoption and productivity;
- continued funding for fundamental computing research, including low-power approaches;
- identifying important UK datasets and testing access arrangements in at least two high-value areas, with health and creative data given as examples.
Keep the frontier models, build some of your own
The fourth recommendation is the most practical for companies selling AI to the British state. The CST does not tell the government to walk away from the largest US models. It says the UK should keep trusted access to the most powerful global models while avoiding reliance on them for every task, because specialised domestic systems may perform better for some jobs.
The pilots would test open-weight models and smaller domestic models against different needs, and the council wants part of the UK's national AI computing capacity set aside for those trials. It also asks ministers to consider a proposal from the Alan Turing Institute for a public-interest sovereign base model: an open-source model built from scratch, released as a public good and governed as national infrastructure for researchers, start-ups, small firms, public services and the open-source community.
What the letter leaves out
The advice carries no price tag. It gives no funding figures and no delivery timetable, and the publication on GOV.UK does not include a formal government response. Ryan Daws, senior editor at AI News, pointed to the same gap in his write-up of the letter on 19 October.
That matters because the hard choices sit in the parts the letter leaves open: how much national compute to reserve for pilots, which datasets are sensitive enough to keep under UK control, and how a sovereign model would be paid for once built. The letter says the council can give further independent advice as needed.
Why it matters beyond Britain
Governments from Paris to Nairobi are asking the same question about dependence on a handful of foreign AI providers. The CST framing, which maps dependencies first and then decides what to own, partner on or rent, is a template other mid-sized economies can borrow. It also sits alongside the UK's regulatory push on AI agents; our report on the ICO's call for evidence on agentic AI sets out what the data regulator is asking developers to explain by 20 November.
For AI vendors, the signal is clear enough. British public-sector contracts may increasingly come with questions about where data sits, who controls the model and whether a domestic or open-weight alternative was tested first. Whether ministers adopt the plan, and fund it, is the next thing to watch.
Source: Council for Science and Technology, "AI sovereignty: preserving UK agency in an AI-enabled world", GOV.UK, published 19 October 2026.
