The UK government has said its artificial intelligence strategy is designed so that Whitehall is not locked into the American AI models it uses today, after figures showed that 10 of the 11 AI tools in central government rely on suppliers from the United States.

The disclosure came in a freedom of information response to UK AI firm Aventail, which revealed that government tools use models from Google, OpenAI, Microsoft and Anthropic, either directly or through Microsoft Azure and Amazon Web Services. The only exception is PolyAI, a British company whose models are used by the government's shared services provider, SSCL.

Tools built to swap models

The Cabinet Office told Computer Weekly that the tools are engineered so the underlying model can be changed. A government spokesperson said tools developed by teams such as the Government Digital Service and the Incubator for Artificial Intelligence, known as i.AI, use decoupled, model-agnostic designs.

"By separating our data pipelines, user interfaces, and retrieval mechanisms from the underlying foundation models, we avoid hardcoding government applications into any single vendor's ecosystem," the spokesperson said.

The spokesperson added that using commercially available models under secure enterprise agreements had let civil servants prototype and deliver "immediate efficiency", while modular API and orchestration layers preserve the ability to evaluate, swap or self-host alternative models as they mature.

The test: what if access were cut off?

Joe Ward, co-founder and chief executive of Aventail, who submitted the freedom of information questions, said model-agnostic design was the right approach and the government deserved credit for thinking about flexibility.

But he added a harder test: "The key test for the British government: if the US president revoked access to every US model overnight, would our AI systems stay afloat? Resilience depends on both adaptability and speed."

The Cabinet Office confirmed that Google Gemini is the primary authorised enterprise-level generative AI tool in central government. Other tools named in the response include Hansard AI, used for parliamentary and legislative research, Minute, which creates meeting notes, and GCS Assist, used for government communications.

The response also said the government has a contingency plan, but it declined to share details, saying disclosure could help criminals plan and execute crimes and avoid detection.

Minister warns against dependence

UK AI minister Kanishka Narayan struck a similar note in a speech last month, stressing that Britain should remain technology agnostic. "We want the world's leading companies to invest here, but we will not accept dependence as the price of that investment," he said.

Narayan warned that much of the technology shaping the economy, public services and national security is developed outside the country, calling the reliance on foreign technology a question of national security and of democratic power. "If the British state cannot exercise meaningful control over the technologies on which our country depends, neither can the people that it represents," he said.

The debate comes as the government pushes ahead with a wider AI package announced around the autumn Budget, including hundreds of millions of pounds for AI security, sovereign AI and AI-driven growth. For now, the official line is that Britain will use the best models available today, while keeping the freedom to walk away from them tomorrow.