TL;DR
CETaS, the security research centre at the Alan Turing Institute, published “Delivering the UK’s AI Future: A Sovereignty Roadmap for 2030” on 18 November. It argues the UK should not try to match US and Chinese frontier labs, but should pool compute, money, data and expertise to build a capable, medium-sized open-weight model of its own, with a first build targeted for late 2027.
The main proposals
The report’s organising idea is “resilience through optionality”: rather than self-sufficiency, the UK should keep credible fallbacks wherever critical dependencies remain, so essential services keep running if access is cut. A home-grown base model would act as that fallback in a crisis and as a foundation for sensitive work in defence, security and the infrastructure the country cannot do without. Organisations would still reach for top frontier models when those clearly do the job better.
On governance, CETaS wants the AI Minister to take charge of sovereignty by early 2027, with a Cabinet sub-committee set up for the purpose. Government would convene industry and civil society behind the model programme, through the AI Taskforce and the Prime Minister’s office, and the Turing says it could take a central technical role.
To keep growing firms in Britain, the report cites DeepMind and Arm as companies founded here that are now foreign-owned. Its answer is a Strategic AI Capability (SAIC) label, reserved for a handful of firms and assets judged critical. Designated firms that make proportionate commitments to keep key capabilities here and invest in UK jobs and infrastructure could, in return, get quicker routes to funding, state compute and trusted datasets, along with simpler routes into public contracts and faster visas for technical hires.
Procurement and switching
For public-sector buyers, CETaS wants resilience weighed with price and performance. Where disruption is a greater risk, critical sectors should hold two separate model options, preferably from vendors based in different countries, and rehearse moving between them. Services that matter most should add a UK-hosted fallback model that has actually been tested.
Report author Sam Stockwell said decisions in the next 12 to 18 months “will determine whether we retain meaningful control” over technologies that underpin the economy and public services.
Looking forward
The roadmap follows CETaS’s September sovereignty vision, and attaches dates to several of its asks. In our view, the dual-supplier recommendation looks the easiest for procurement teams to act on, because it changes how services buy rather than requiring something new to be built. The shared sovereign model is the most ambitious idea, and it needs government to back the programme the report proposes. The same week, a Commons committee called for statutory frontier AI safety rules, so sovereignty and safety are now being argued in parallel.