Britain cannot let artificial intelligence be ruled by “tech bros and oligarchs”, the UK’s AI minister Kanishka Narayan told the BBC’s Political Thinking podcast — arguing that handling the technology requires not only regulation but “building capability and capacity in the state”.
The number in the middle of it
Narayan said the government recently used frontier AI models in Whitehall to check for cyber security issues. “We only spent £13,000 on it,” he said — and the exercise surfaced 500 new vulnerabilities in government systems.
That figure deserves more attention than the soundbite around it. Thirteen thousand pounds is roughly the price of a second-hand car. If a national government can run an AI-assisted security audit across its own estate for that, then cost has stopped being a reason for anyone not to — and the question becomes why it had not been done already.
The argument he is making
Narayan, appointed the UK’s first cabinet minister for AI in July, compared the scale of change to the Industrial Revolution and said it “in some ways exceeds” it. His warning is about the transition rather than the destination: in the Industrial Revolution “the outcome as a whole was good for the country, but the transition was extremely painful for too many people”.
“The fundamental thing is that the British public and democracy has to assert power rather than leave it to tech bros and oligarchs,” he said, naming no companies. Beyond “pulling the lever of regulation”, he argued Britain needs to build AI domestically “so you can shape the working conditions”, adding: “we’re now currently totally reliant on other countries for essential bits of technology.”
His comments were recorded the day before Barclays announced it was cutting 5,000 jobs in an AI-led efficiency drive, and before the PHP flaw — found with AI’s help — became public.
The sentence that applies here without translation
“We’re now currently totally reliant on other countries for essential bits of technology.” That is a British minister describing Britain. It is also, with no adjustment at all, a description of Bangladesh — where the cloud is foreign, the models are foreign, the chips are foreign, and now the language runtime that half the country’s websites depend on turned out to need an emergency patch written somewhere else.
The transferable part is not the rhetoric. It is the £13,000. Bangladesh has a national CIRT, dozens of ministries with public-facing systems, and no shortage of people who could run an AI-assisted audit across them. The barrier has been assumed to be money. This week’s figure says otherwise, and the honest obstacle is that nobody has been told to do it.
What the £13,000 does not cover
Finding 500 vulnerabilities is the cheap half. Triaging them, deciding which are real, assigning owners, scheduling fixes across departments that each control their own systems, and verifying the work afterwards — that is the expensive, slow, unglamorous part, and no model does it for you.
This is worth stating because the headline invites the wrong conclusion. An audit that produces a list nobody acts on is not security; it is a document. The useful metric from the Whitehall exercise will not be how many issues AI found for £13,000. It will be how many were closed, and how long each took.
