UK AI minister tells frontier labs to do more on safety

The UK's AI minister, Kanishka Narayan, will tell companies building the most advanced AI systems that they must increase their safety efforts, in a intervention that comes alongside a strengthening of Britain's cyber defences.

The message, first reported by The Guardian, marks a sharpening of tone from the British government toward the developers of so-called frontier AI — the most capable models in the industry. Narayan will argue that Britain cannot rely entirely on technology companies to police their own models, according to the report.

What is the minister saying?

Narayan's remarks, as previewed, contain three distinct strands:

• A direct call on companies developing frontier AI "to do more to make sure it is safe."
• A warning that the UK "cannot rely entirely on technology companies" to police their own models.
• An argument that the UK "needs to build more of its own AI infrastructure."

The intervention arrives amid what The Guardian characterises as rising global concern over AI dangers — a backdrop that has pushed AI safety from a niche policy debate into mainstream regulatory discussion across major economies.

Why does this matter?

The stance has stakes on two fronts.

First, it signals where the UK stands on a question every government deploying advanced AI now faces: whether voluntary commitments from labs are sufficient, or whether the state must build its own capacity to evaluate and defend against AI risks. Narayan's framing — that reliance on companies alone is not enough — positions the UK closer to the interventionist end of that debate.

Second, the announcement of strengthened cyber defences ties AI policy directly to national security. Frontier models are increasingly seen as both a target for attacks on critical infrastructure and a tool that hostile actors can weaponise, which makes defensive capability a precondition for any credible AI strategy.

What does the call for domestic AI infrastructure imply?

Narayan's argument that Britain needs to build more of its own AI infrastructure speaks to a long-running concern in UK industrial policy: that the country hosts world-class AI research but owns too little of the compute, data-centre and engineering base on which the technology runs.

Dependence on foreign-built and foreign-operated infrastructure limits how a state can independently test, audit and secure the AI systems its economy depends on. A domestic capability would give the government the technical means to back its safety demands — something that oversight based purely on corporate self-reporting cannot provide.

What comes next?

The full speech, expected imminently, will fill in the detail behind the previewed remarks — including what "strengthened cyber defences" concretely involves and whether the government pairs its call for corporate action with new regulatory or investment measures. The direction of travel, though, is clear: the UK is signalling that frontier AI safety is now a matter of national capability, not just corporate responsibility.