The UK government is moving responsibility for artificial intelligence closer to the prime minister, concentrating public-sector adoption, policy coordination, and parts of the country’s AI security machinery inside a new central taskforce.

Patrick Vallance will chair the body within the Office for the Prime Minister and the Cabinet. Minister for AI and Online Safety Kanishka Narayan will lead its political work and attend Cabinet, while the taskforce will report directly to the prime minister.

The AI Security Institute will also move into the new structure, bringing research into advanced model risks under a more politically central organisation. Vallance will retain responsibility for the Oxford-Cambridge Growth Corridor, connecting the role with infrastructure, research, housing, and regional development.

Artificial intelligence will consequently gain unusual proximity to executive power as departments are encouraged to use it in administration and public-service delivery. The change also adds another institutional layer to a government technology system already divided across departments, the Government Digital Service, commercial functions, regulators, and specialist agencies.

Central authority can remove organisational barriers

Public-sector AI projects often stall between departmental ownership and cross-government requirements. Individual organisations control budgets and operational data, while central bodies set technology standards, security rules, procurement frameworks, and spending controls.

A taskforce reporting to the prime minister may be able to resolve disputes or force cooperation that would otherwise move slowly. Such authority could prove useful where adoption depends on shared identity systems, data access, cloud services, model evaluation, and procurement arrangements that cannot be recreated efficiently by each department.

Central coordination may also give ministers a clearer view of which pilots are producing usable results and which are consuming money without reaching operational deployment. However, authority cannot substitute for the staff and technical capability required to redesign a service.

Benefits, health, justice, taxation, local government, and regulation all operate under different legislation and risk. Departments still need engineers, data specialists, service designers, user researchers, commercial expertise, and managers who understand how work will change around the technology.

The taskforce will therefore need to reduce friction rather than creating another approval layer. Its usefulness will depend on whether it simplifies decisions, provides reusable capabilities, and publishes enough evidence to distinguish service improvements from activity labelled as AI.

Adoption and security now sit closer together

Moving the AI Security Institute nearer to the prime minister may increase its influence over policy where model risk intersects with national security. It also raises questions about independence, transparency, and the institute’s relationship with departments responsible for science, technology, cyber policy, and regulation.

AI adoption and AI safety overlap, but their incentives are not identical. A team promoting wider use may prioritise speed and measurable efficiency, while evaluators need the authority to identify systems that are unreliable, insecure, or unsuitable. Clear mandates and published evaluation methods become more important when those functions share an organisational home.

Vallance brings experience at the boundary between science and government, having served as chief scientific adviser during the pandemic and later as science minister. His appointment indicates that ministers increasingly regard AI as an issue of state capacity and economic policy rather than a programme confined to digital teams.

The Oxford-Cambridge connection reinforces that approach. AI growth depends on laboratories, compute, electricity, transport, housing, university research, and private capital, which makes it an industrial and planning challenge as well as a software deployment programme.

Inside Whitehall, progress will be constrained by legacy systems, procurement cycles, inconsistent data, and public-law obligations. Faster document preparation or administrative triage may offer early gains, while systems affecting eligibility, enforcement, healthcare, or individual rights require deeper assurance and routes for challenge.

Centralising responsibility creates a clearer line of authority at the top of government. It also makes the centre more directly accountable for the evidence, safeguards, and service outcomes produced when AI moves from political priority into everyday administration.