The government's own science advisers warn the UK risks becoming an 'AI customer, not a creator'

The UK government’s top independent science advisers have delivered a blunt verdict on the country’s AI position: without “deliberate and urgent action,” Britain risks becoming “a high-value AI customer, rather than a country able to shape the technology and its usage.” The warning comes from the Council for Science and Technology (CST), in a letter to AI minister Kanishka Narayan dated 30 July and published by the government on 19 October.

It is advice, not policy — but it carries weight. The CST is co-chaired by the Government Chief Scientific Adviser, Dame Angela McLean, and Lord Browne of Madingley, and the letter was copied to the Chancellor, the Defence Secretary and the Prime Minister. When the state’s own scientific establishment warns that the national strategy is drifting towards dependence, that is not an outside critic; it is the house view.

What they actually recommend

Importantly, the council is not arguing for AI autarky. Its line is that “while access to the most powerful global models is important, the UK should not depend on them for all use cases” — targeted sovereignty, not isolation. Its four recommendations are practical:

• Map the UK’s AI infrastructure — understand what capability the country actually has and where it is dangerously reliant on others.
• Appoint a single official to coordinate investment decisions, rather than spreading them across departments.
• Strengthen domestic skills, research and access to data.
• Test British AI models in public services such as healthcare and national security, allocating national supercomputing capacity to the trials — and consider the open-source British model proposed by the Alan Turing Institute.

Why it bites this week

The warning would be abstract were it not for the timing. It arrives the same week Microsoft reportedly walked away from Nscale’s Loughton supercomputer — the UK’s largest planned AI facility — over a site that cannot be powered until the 2030s. And it comes barely a month after the UK’s AI Security Institute had its access to US frontier models switched off for a fortnight at Washington’s discretion. The dependency the advisers describe is not hypothetical; those two events are what it looks like in practice.

The government is not doing nothing — its sovereign-AI programme is taking equity stakes and buying compute from British companies, exactly the “targeted domestic investment” the CST calls for. But the council’s framing is useful because it names the choice plainly: for which uses does the UK need its own frontier-model capability, its own compute, its own data — and for which is renting from the US and relying on goodwill genuinely fine? That is the question the whole sovereign-AI debate, which runs through our coverage and the government’s, keeps circling. Its advisers have now put it in writing.