The artificial intelligence (AI) tools used by the UK government rely almost exclusively on models from US tech firms, raising questions over contingency planning.
According to a response to a freedom of information (FOI) request from UK AI firm Aventail, 10 of the 11 AI tools deployed or bought by the Cabinet Office use AI models developed by US tech firms.
While the models are hosted in the UK, the design, access and pricing are controlled by US organisations.
Google, OpenAI, Microsoft or Anthropic are used directly or through Microsoft Azure and Amazon Web Services (AWS). The only exception is PolyAI, a UK company whose models are used for the government’s shared services provider, SSCL.
The Cabinet Office confirmed that Google Gemini is the primary authorised enterprise-level generative AI tool. Other AI tools used include Hansard AI for parliamentary and legislative research, Minute for creating meeting notes, and GCS Assist for government communications.
Joe Ward, CEO of Aventail, the company that made the FOI request, said: “Vital tools at the heart of government depend almost entirely on a handful of US companies. Hosting the data in the UK helps, but it does not change who controls the models, the pricing or the terms of access.”
We need flexible infrastructure that lets public bodies switch providers, adopt new models and run open-weight alternatives under their own control when needed Joe Ward, Aventail
He said he supported the UK government’s use of the latest US models, but added: “We risk building a house of cards if essential services depend on continued access to any one supplier. We need flexible infrastructure that lets public bodies switch providers, adopt new models and run open-weight alternatives under their own control when needed.”
Continuity question
The FOI request to the Home Office, which manages the government’s AI and technology supplier relationships, asked for details of any business continuity plans, risk assessments or risk register entries that address an outage of, or loss of access to, those providers’ services.
In response, the government said the information does exist, but that disclosing it would prejudice the prevention or detection of crime.
“The information contains details about the detection and prevention of crime which would assist criminals to plan and execute criminal acts and to avoid detection. Disclosure would thus make the prevention of certain crimes more difficult and reduce the effectiveness of the police service in detecting the perpetrators,” the Cabinet Office said in its response.
Mark Lewis, a specialist IT outsourcing lawyer at Stephenson Harwood, asked how the government thinks it can “ever extricate itself”.
“It’s not just that they’re all US, but there is a high concentration of the leading developers. Moreover, the infrastructure, of course, is inevitably US cloud,” said Lewis. “It’s all very well talking about sovereignty, but how realistic is it, given that proportion of dependency on not just a handful of US frontier AI labs, but also US cloud?”
He added that supplier lock-in is also a risk that could repeat what has happened in cloud computing: “AI labs locking in customers is exactly the same as we saw with the cloud.”
But Lewis added that the government could well have negated risks through the deals it signed. “What we don’t know, in fairness to the UK government, is whether they’ve managed to do smart commercial deals. But, bluntly, given what we know about UK government procurement – and in particular IT procurement – it’s hard to imagine that the UK government would be in a strong position. But we don’t know, in fairness.”
Table of Contents CJEU ruling focus on evidence sharing between EU states Defence secrecy Investigations began in 2018 German courts disagree on law CJEU decides…
Corporate IT teams have expressed discomfort towards the use of employee surveillance technologies at work, with both workers and managers highlighting its potential to hasten…
The UK’s National Cyber Security Centre (NCSC) has rolled out a series of updates to its Cyber Assessment Framework (CAF) aimed at assisting operators of…
Generative AI (GenAI) tools used by cyber criminal fraudsters to create convincing deepfake voice scams appear to have mastered the fine art of the Glaswegian…
Table of Contents Only a small proportion of summaries checked by experts Human in the loop? More transparency needed over AI An artificial intelligence (AI)…
In search of the easiest route to a quick pay-off, ransomware gangs in 2023 have been increasingly targeting small and medium-sized enterprises (SMEs) with under…
