Britain's AI policy is a language problem before it's a policy problem

Neville Hobson 06 Jul 2026 4 min read

Back in May, I wrote about Tony Blair's essay on the future of the Labour Party and the country, and one line stuck with me – Blair's argument that governing in the age of AI will be the defining challenge for any government in the foreseeable future.

It was a compelling notion when I wrote it. It's rather more than that now.

That essay landed before Keir Starmer's resignation and the rapid ascent of Andy Burnham as his likely successor. And last week, the Financial Times reported that Burnham's team is planning to revamp the UK's AI strategy – pivoting away from what they characterise as a US-centric model pushed by former Washington ambassador Peter Mandelson, and towards British companies and workers instead.

I don't want to write about which of these approaches is right. What I find genuinely interesting – as someone who spends a lot of time thinking about how communicators frame things – is that we can now watch two rival vocabularies for talking about AI compete inside the same governing party, before either one has been tested against reality. And the vocabulary is doing a lot of the work here, well ahead of any actual policy.

Two registers, side by side

The Mandelson-era register is the language of partnership and scale: investment deals, the £22bn Microsoft commitment, £5bn from Google, Sam Altman described as Mandelson's "chief AI buddy." It's a growth story, told in the currency of big numbers and big names.

Burnham's team is speaking a different language entirely. "Unfettered tech boosterism is a vote-loser," one adviser told the FT. The current government's courtship of US tech is described not as investment but as a "geopolitical failure." The recurring question, repeated more than once in the piece, is disarmingly simple: "what's the point and who's it for?"

That's not a left-right split, incidentally – it's a fault line inside one party, which makes it a cleaner case study in framing than most political arguments, where the underlying tribal loyalties usually do half the persuading for you.

The rhetorical moves worth noticing

A few things stand out if you read the coverage as a communicator rather than as a political observer:

Investment reframed as risk. US commitments that were, until recently, presented as a win are recast as dependency – one contributor to the piece invokes the idea of Britain becoming a "vassal state." Same facts, opposite emotional charge.
The folksy challenge question. "What's the point and who's it for?" is doing something specific – it takes a complex infrastructure and procurement decision and turns it into a fairness question anyone can have a view on.
Naming real people to make the abstract concrete. Black cab drivers and Uber drivers appear by name in the reporting on driverless cars – a deliberate move to give a policy shift a human face before it's even been decided.
A coined political philosophy. "Manchesterism" isn't just a policy platform, it's a brand for a governing style, introduced in a speech rather than a white paper.
What's largely absent. Neither register borrows much from the AI industry's own vocabulary – GEO, agentic systems, frontier models, the language you'll have heard if you follow the FIR podcast or my own posts on this. That's telling in itself. The political conversation and the technical one are barely speaking to each other yet.

Why this matters beyond politics

This is where it connects back to communicators more broadly, not just political watchers. What's happening at Westminster – building the emotional case for a policy before the technical case is settled – is exactly what plenty of organisations are quietly doing internally with their own AI messaging.

An efficiency narrative versus a workforce-anxiety narrative. A sovereignty-and-control story versus a partnership-and-scale one. The dynamics are the same size, just smaller stakes.

There's also a real-world anchor worth mentioning here, and I'll flag that some of the detail below is still developing. The FT piece notes that Burnham's team is leaning on tech sovereignty partly because of the US government's export controls on Anthropic last month, which forced Anthropic to withdraw its Mythos and Fable models for users worldwide before the controls were lifted.

Whatever you make of that episode, it's a useful demonstration that "sovereignty" arguments about AI aren't purely rhetorical – they can become operational very quickly, and affect real users overnight.

Back to Blair's "defining challenge"

Read together, I think the FT piece is actually evidence for Blair's underlying point, even though it complicates his preferred answer to it. He framed AI governance as a challenge to be met with vision and partnership. What we're watching instead is that same governing party reaching for two entirely different vocabularies to describe what "meeting the challenge" even means – within months of his essay, not years.

That, to me, is the real story here. Not who wins the internal argument, but how quickly AI has become significant enough to reshape the language a governing party uses to talk about itself.

A question worth sitting with

I'm not going to offer a verdict on Mandelson's approach versus Burnham's – that's a job for people closer to the policy detail than I am, and there are reasonable arguments on both sides that deserve more space than a blog post allows.

But if you work in communications, it might be worth a few minutes noticing which of these two vocabularies – partnership-and-scale, or sovereignty-and-protection – is showing up in your own organisation's AI messaging. Watching how people choose to talk about AI, right now, tells you more about where opinion is heading than any strategy document will.

Sources: