The UK will establish a national AI Biosecurity Lab to strengthen its defences against emerging biological threats, committing £100m in Wednesday’s Budget to what the government calls a “world first” institute.
What was announced
Alongside the lab, the chancellor pledged a further £120m for the UK’s AI Security Institute, established in 2023 to evaluate the risks of the most advanced models. Millions more are promised across initiatives aimed at cyber, biological and AI-enabled threats.
“AI will transform daily life: every firm, community, hospital, school, and every aspect of our economy can benefit,” John Healey said. “I understand and share the concerns of working people, so I am setting out a distinctively British approach to AI. One that doesn’t leave workers behind and defends our national interest.”
AI minister Kanishka Narayan described the lab as “a major new global institution” bringing together “world-leading scientific expertise”, with the aim of “staying ahead of emerging threats, from cyber attacks to biological risks, while building more of the critical technology our security depends on”.
The full remit is not yet public.
The funding detail worth pausing on
OpenAI and Anthropic will support the lab by providing £200m worth of AI model credits — that is, compute that can only be spent on their own models.
Read that twice. A state institution established to assess biological risk, much of it arising from AI systems, is being part-resourced by two of the companies building those systems, in a currency redeemable only with them. It is not a scandal and it may well be the fastest way to get the lab working. But it is a dependency, and dependencies shape what gets examined. An institute whose compute budget is a donation from two vendors has a structural reason not to make their models its first case study.
Why a small country should read this closely
Because the same offer arrives everywhere, in smaller denominations. Model credits, free tiers for government, donated infrastructure for a university programme — these are the normal instruments by which AI companies enter a national market, and they are genuinely useful. Bangladesh is drafting its own National AI Policy for 2026–2030, and that document will eventually have to say something about accepting resources from the firms it intends to regulate.
The useful question is not whether to accept. It is what independence costs and who pays for it: if the credits run out, does the capability stop? If the answer is yes, the institution is a customer rather than a regulator, whatever the founding statement says.
Why biosecurity, and why now
The specific fear is not that an AI model builds a pathogen. It is that a capable general model lowers the expertise required to do something that previously needed a trained specialist — turning a problem of knowledge into a problem of access. That is why evaluation work in this area concentrates on what a model will explain to a determined non-expert, rather than on what it can do by itself.
The timing is therefore as much a political choice as a scientific one. Announcing a biosecurity institute in a Budget, alongside £120m for the AI Security Institute, lets a government say it is acting on AI risk without regulating any product. Building capability is genuinely useful and considerably easier to legislate than restraint.
