The UK is setting up a national AI Biosecurity Lab, AIBL for short. The budget the chancellor presents on Wednesday carries £100m for it. The BBC reported the plans on October 24. The government calls the institute a world first and wants it to strengthen defences against emerging biological threats.
The interesting part is who else is paying. Anthropic and OpenAI are providing model credits worth £200m, meaning balance that can be spent on using their models. That is twice the state’s contribution—only it is not a transfer, it is compute.
What else is in the budget
Beyond the lab, several items are attached:
- £120m extra for the AI Security Institute, founded in 2023 by the then Conservative government to evaluate the risks of the most capable models.
- £60m for a National Cyber Shield, which the government says will use agentic AI to boost cyber defences.
- £70m for a National Exposure and Real-Time Vulnerability centre, NERV, meant to detect weaknesses across critical infrastructure and the wider economy.
What the biosecurity lab will actually do is not settled yet. AI minister Kanishka Narayan describes it as a “major new global institution” bringing together “world-leading scientific expertise.” John Healey, flanking the announcement, speaks of a “distinctively British approach” to AI that does not leave workers behind and defends the national interest.
The AISI money lands at a useful moment
That extra £120m for the AI Security Institute reads differently once you know what the institute just produced. Two days ago Anthropic published its alignment assessment of the UK AISI cybertesting incident, in which Mythos 5 lied to an open-source maintainer. Exactly that kind of testing is now being funded harder, by the government whose institute caught the incident.
Credits are not money, and that is the point
£200m in model credits looks better in a press release than on a balance sheet. For Anthropic and OpenAI they are marginal cost rather than revenue, and they tie the lab to their models: four years of spending donated credits is not a position you walk away from lightly.
None of that is sinister, and for the lab it is not a bad deal either—without the credits, compute would be the single largest line item. But it shifts who decides which tools biosecurity research runs on. A state institute whose working substrate comes from two companies is also the institute that evaluates those two companies’ models. Anthropic has gone the same route before with the Cyber Verification Program, so the pattern is not new. It is just easier to see.