TL;DR

Kanishka Narayan, the UK’s cabinet minister for AI, has said regulation alone will not be enough and that the state must develop its own AI capability. In an interview with Nick Robinson for Political Thinking, the BBC podcast, published on 23 October, he said a frontier-model security check of Whitehall systems costing £13,000 had turned up 500 previously unknown vulnerabilities. He also hinted at an AI measure in next week’s Budget.

Power, not just rules

Narayan framed AI as a change on the scale of the Industrial Revolution, which he said ended well for the country overall but was “extremely painful for too many people” along the way. His answer is for “the British public and democracy” to “assert power rather than leave it to tech bros and oligarchs”. No firms or executives were named.

Alongside “pulling the lever of regulation”, he argued the UK should build AI at home so it can influence working conditions, and to cut what he called a total dependence on foreign suppliers for core technology.

The £13,000 test

The most concrete claim in the interview concerned government security. “We only spent £13,000 on it - we discovered 500 new vulnerabilities,” Narayan said of the frontier-model exercise, describing them as novel flaws in significant parts of the public sector. He gave no further detail.

The conversation was taped a day ahead of Barclays’ 22 October announcement of around 5,000 job cuts, and before the AI-found PHP flaws that prompted UK government checks. He also rejected the idea that only the US and China matter, saying “every US lab” had come to the UK’s AI Security Institute asking for models to be tested before release.

Looking forward

Narayan said talks with Chancellor John Healey pointed to a Budget measure on AI next week, without saying what it might be. That is the next place to look for whether building state capability comes with money attached. This interview follows his warning earlier this month that a breach of UK cyber defences is likely. In our view, the £13,000 figure is the claim most worth pinning down: if a small spend on frontier models can surface hundreds of public-sector flaws, the obvious follow-up is how quickly those flaws are being fixed.