PolicyLaw, regulation & governance
UK government to base its AI policy on 'control and sovereignty' principles
Britain's government will build its AI policy around 'control and sovereignty', the FT reports. Over recent months, ministers recast AI as a security issue, government advisers urged a sovereignty plan and the Budget set aside £700m for sovereign capability and security.
From the Financial Times · AI article of November 2, 2026 and cited sources: According to the FT, 'control and sovereignty' will guide how the UK rolls out AI. [1]
SOURCES[1] ‘Control and sovereignty’ to shape UK approach to AI ft.com
AT Pulse research deskPublished · Updated · 5 min read · AI-assisted analysisAI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- According to the FT, 'control and sovereignty' will guide how the UK rolls out AI. [1]
- Original source
- Financial Times · AI article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 2, 2026 article published
- Covered
- Nov 2 edition
- Evidence
- 18 sources across 15 sites
- 1‘Control and sovereignty’ to shape UK approach to AIft.com
- 2Government Technology: Kendall speechgovernmenttechnology.co.uk
- 3City AM: £500m Sovereign AI fundcityam.com
- 4Bramble Hub: Kendall speechbramblehub.co.uk
- 5The Next Web: Cosine Lumen Sovereignthenextweb.com
- 6GOV.UK: AI sovereignty: preserving UK agency (CST)gov.uk
Reach: not measured yet
According to the FT, 'control and sovereignty' will guide how the UK rolls out AI. The principles came out of high-level debate inside government, and their wording has not been released. [1]
Funded policy steps came first. In April, Technology Secretary Liz Kendall described control over AI as a national-security matter, and a £500m Sovereign AI Unit was launched. In October, the government's Council for Science and Technology called for a UK AI Sovereignty Plan. The Budget then reportedly allocated £700m, including £140m for the Sovereign AI Unit, £120m for the AI Security Institute and £40m to investigate incidents involving autonomous AI agents. [2] [3] [6] [7] [8]
Official and expert sources define sovereignty as having enough say over chosen parts of the AI supply chain, not as self-sufficiency. A paper from CETAS, the Turing Institute's security centre, separates four kinds of control: contractual, technical, possession of model weights and operational. Each needs different engineering. The Council also advises keeping trusted access to leading foreign models. [6] [10] [11]
The direction and funding are documented, but how the principles will apply is not. If public buyers adopt CETAS's control types, they may ask where models run, who holds the weights and whether audit access is guaranteed.
Read the full assessment
That would favour UK-hosted and on-premises vendors.
Executive brief
The UK government plans to build its AI approach around "control and sovereignty". The Financial Times reports that these principles followed "robust discussions" at the top of government. The FT's actual wording of the principles could not be checked against any other source. What can be checked is the policy trail behind them. Over 2026, ministers redefined AI as a national-security issue, a government advisory council called for an AI sovereignty plan, and the 28 October Budget set aside £700m for sovereign capability and security. The sovereign push is real and funded. How the new principles will be applied has not yet been published.
What changed and event timeline
£500m Sovereign AI Unit launched
The new unit is meant to invest in British AI start-ups and scale-ups, with Kendall saying she would back "British winners".
More detail
Kendall reframes AI as security
In a speech at the defence think tank RUSI, Technology Secretary Liz Kendall said control over where AI is built and who controls it underpins economic, energy and defence security. She rejected isolationism and promised a UK AI hardware plan.
Industry coalition for a "sovereign" model
The start-up Cosine announced Lumen Sovereign, to be trained on the Isambard-AI supercomputer. BT, HSBC, Lloyds, NatWest and BAE Systems are among the named backers, but under memoranda of understanding rather than contracts.
Advisory council's AI sovereignty letter published
The Council for Science and Technology (CST) made four recommendations: a UK AI Sovereignty Plan, clear governance, investment in foundations, and continued access to leading models. The letter was sent to the AI minister in July.
Budget commits £700m
The package includes £140m for the Sovereign AI Unit, £120m for the AI Security Institute and £40m to investigate incidents involving autonomous AI agents.
"Control and sovereignty" principles reported
The FT says the principles came out of high-level internal debate over how to roll out the technology. The text of the principles has not been published.
Capabilities and access
This is a policy story, so no new government model or version is involved. The capability most closely tied to it is Cosine's Lumen Sovereign:
Read the full section
This is a policy story, so no new government model or version is involved. The capability most closely tied to it is Cosine's Lumen Sovereign:
- Cosine says it is being built from scratch, not fine-tuned from an open model, using proprietary data covering 30+ regulated workflows.
- It is meant to run on air-gapped (fully offline) systems and to be delivered by the end of 2026. The Next Web
- Training compute is allocated through the government's Sovereign AI programme. CreditStrategy
- Model size, architecture and benchmark results have not been disclosed.
Technical analysis for researchers and developers
- The CST frames sovereignty as "sufficient agency across the AI value chain", not self-sufficiency.
- A paper from the Turing Institute's security centre (CETAS) separates them: contractual, technical, possession of model weights, and operational.
- The £120m for the AI Security Institute and £40m for agent-incident work point to more state-run testing. No methods or protocols have been published.
Read the full section
- Definition. The CST frames sovereignty as "sufficient agency across the AI value chain", not self-sufficiency. Under that definition, "control" means control over specific layers of the stack. GOV.UK
- Four kinds of control. A paper from the Turing Institute's security centre (CETAS) separates them: contractual, technical, possession of model weights, and operational. Each needs different engineering, for example on-premises deployment, key custody or audit access. CETAS
- Evaluation. The £120m for the AI Security Institute and £40m for agent-incident work point to more state-run testing. No methods or protocols have been published.
Claims and evidence
- The principles stem from "robust" top-level debate. This comes only from the FT's standfirst and has no other corroboration. FT
- £700m Budget package and its breakdown. Reported by trade press, but not confirmed against Treasury documents. AI Magazine
- £60m added to the Sovereign AI Fund.
Read the full section
- The principles stem from "robust" top-level debate. This comes only from the FT's standfirst and has no other corroboration. FT
- £700m Budget package and its breakdown. Reported by trade press, but not confirmed against Treasury documents. AI Magazine
- £60m added to the Sovereign AI Fund. Matches two separate accounts. The Conversation via NewSpace Economy
- Cosine's coding models beat OpenAI, Anthropic and others. This comes from Cosine itself, and no benchmark figures were published with it. The Next Web
Context and prior work
- Cambridge's AI centre published a set of strategic choices on UK AI sovereignty in October 2025. Cambridge CETAS has also set out a vision for 2030. CETAS
- Academics Vincent Charles and Tatiana Gherman count 184 sovereign-AI projects across 67 countries by mid-2026. They note that about three-fifths of the foreign base models they tracked come from the US. NewSpace Economy
Read the full section
- Earlier frameworks. Cambridge's AI centre published a set of strategic choices on UK AI sovereignty in October 2025. Cambridge CETAS has also set out a vision for 2030. CETAS
- Global trend. Academics Vincent Charles and Tatiana Gherman count 184 sovereign-AI projects across 67 countries by mid-2026. They note that about three-fifths of the foreign base models they tracked come from the US. NewSpace Economy
Limitations, safety and contested findings
- MPs warned in July that the UK has no coherent sovereign-AI strategy. ResultSense
- The industry body UKAI has questioned whether the Budget is big enough. ResultSense Tech executives have called for £20bn of investment. ITPro
- What sovereignty can mean. Charles and Gherman argue that sovereignty means managing dependencies, not eliminating them. NewSpace Economy
Read the full section
- Coherence. MPs warned in July that the UK has no coherent sovereign-AI strategy. ResultSense
- Scale. The industry body UKAI has questioned whether the Budget is big enough. ResultSense Tech executives have called for £20bn of investment. ITPro
- What sovereignty can mean. Charles and Gherman argue that sovereignty means managing dependencies, not eliminating them. NewSpace Economy
- Firmness of industry support. The Lumen Sovereign backers have signed non-binding memoranda only.
Business and practitioner implications
- If public-sector buying follows the principles, expect questions about where AI runs, who holds the model weights, and whether audit access is guaranteed.
- Air-gapped and on-premises deployment, UK hosting and contractual guarantees may become competitive advantages.
- Early-stage UK AI firms can now apply to dedicated money: the Sovereign AI Fund, the AI Pathfinder fund and local-partnership pilots. Technology Magazine
Read the full section
- Procurement. If public-sector buying follows the principles, expect questions about where AI runs, who holds the model weights, and whether audit access is guaranteed. These are the CETAS control types.
- Vendors. Air-gapped and on-premises deployment, UK hosting and contractual guarantees may become competitive advantages.
- Funding. Early-stage UK AI firms can now apply to dedicated money: the Sovereign AI Fund, the AI Pathfinder fund and local-partnership pilots. Technology Magazine
- US frontier labs. The CST still recommends keeping trusted access to leading models, which suggests partnership terms will be negotiated rather than replaced.
The source trail.
Sources (18)
‘Control and sovereignty’ to shape UK approach to AI
Feed excerpt only.
ft.com