Does the UK AI Security Institute have sufficient powers to protect against AI harms?

The story here is consensus, not conflict: when a body can evaluate but not compel, can warn but not block, voters of every generation reached for the same word — insufficient. Sixteen said the powers were adequate against forty-one who didn't, and that ratio held with unusual steadiness from the youngest cohort to the over-65s, a rare cross-generational agreement in a category, AI governance, usually defined by generational anxiety gaps.

If there's a fracture, it's mild and it's gendered: women voters split closer to evenly, while men were more lopsidedly sceptical, suggesting the harder line came less from fear of AI itself than from a sharper read on regulatory teeth versus regulatory theatre. But with only eight women voting, that's a texture, not a trend.

The result lands squarely on the fault line running through UK tech policy since the Bletchley process began — a preference for "pro-innovation" light-touch oversight now colliding with public appetite for enforceable red lines, the same tension that has dogged Ofcom's online safety powers and the AI Safety Institute's voluntary model-testing arrangements with industry.

An institute built to advise, not to arm-wrestle Silicon Valley, was never going to satisfy voters looking for a regulator with teeth — and on this evidence, it hasn't.