UK’s ‘sovereign AI’ overwhelmingly dependent on US tech

Ten of 11 AI tools disclosed by the Cabinet Office rely on models from US companies, despite government efforts to build sovereign AI capability

The ideal of “sovereign AI” has featured prominently in several recent ministerial speeches, but a freedom of information (FoI) request to the Cabinet Office by UK AI inference start-up Aventail reveals the gulf between that goal and the reality.

Of 11 AI tools built, deployed or procured by Government Digital Service (GDS) and the Incubator for Artificial Intelligence (i.AI), the government’s in-house AI development unit, ten depend on models made by OpenAI, Anthropic, Google or Microsoft, with several deployed through Microsoft Azure or Amazon’s AWS cloud services.

The only exception is UK company PolyAI, whose conversational AI technology is used by government shared services provider SSCL to power AI voice assistants for contact centres.

The AI-powered applications based on US models include the internal UK government generative AI drafting assistant Gov Draft (“typically OpenAI and/or Anthropic via Microsoft Azure or Amazon Web Services Bedrock”, according to the Cabinet Office); parliamentary and legislative research tools Parlex and Lex (Microsoft/OpenAI and Anthropic); GCS Assist for government communications (Anthropic); the Fraud Risk Assessment Accelerator (Microsoft/OpenAI); and Minute for meeting notes (Microsoft Azure OpenAI, AWS and Google Gemini).

The government has committed more than £1 billion to expanding domestic AI compute capacity and wants to increase UK AI compute twentyfold by 2030. However, as Computing has repeatedly reported, sovereignty is more than simply locating infrastructure or data in UK data centres.

Google Gemini, the Cabinet Office's primary authorised enterprise generative AI tool, operates within what the department calls "UK or EU-sovereign boundaries". However, the models remain under Google's control.

Recent Computing research found a similar contradiction in other public and private sector organisations: 87% of 100 UK IT leaders said digital sovereignty was important to their cloud strategy, including 20% who considered it mission-critical, but relatively few were actively migrating away from US or Chinese providers.

Not that it's an all-or-nothing decision.

Former GDS head Mike Bracken suggests that complete technological independence is neither achievable nor necessarily desirable for the UK. Nevertheless, he argues that the authorities should be actively considering which capabilities the country needs to control and which dependencies it is willing to accept rather than simply rolling with the default options.

The Cabinet Office appears conscious of its dependency on overseas model suppliers. In the FoI response a spokesperson confirmed that the department has business continuity plans and risk assessments that cover the loss of access to its AI providers. However, officials declined to provide details on law-enforcement grounds.

Aventail CEO Joe Ward described the situation as “precarious” and “an unstable house of cards that’s one geopolitical fracas away from collapse”.

Ward continued: “The AI capabilities of the British state are almost entirely stacked on a handful of American companies.”

He added: “The solution can only be flexible infrastructure: the ability to easily connect to different models, including open-source alternatives that are under their own control.”