£600m and a Biosecurity Lab: The Government's AI Spending Spree Nobody Can Actually Account For

Big Numbers, Fuzzy Maths

I've been in enough pitch rooms to recognise a headline number when I see one. Chancellor John Healey's £600m AI security package, dropped on Saturday morning four days before the Budget, is a masterclass in the genre. Spread across enough initiatives and it photographs beautifully. But does it hold together under scrutiny?

The UK government set out around £600m of funding, largely geared towards defending against health and security threats from AI. Reuters confirmed the figure the same morning it landed. The centrepiece is the AI Biosecurity Lab, which the Cabinet Office is calling, without a flicker of self-consciousness, the world's first dedicated AI biosecurity organisation. The lab gets £100m in public money, and up to £200m in model credits from OpenAI and Anthropic.

Pause there. The OpenAI and Anthropic commitment is in model credits, not cash. The Cabinet Office confirmed that OpenAI will provide £100m in compute and model credits through its Rosalind Biodefense programme, and Anthropic another £100m in model credits and technical support. That is not another £200m in public cash funding. Worth stating plainly, because the early headlines weren't always plain about it.

So the actual government outlay is around £600m, which is still enormous. The lab's total package, mixing public funding with company credits whose value depends on how much they are actually used, is being reported by some outlets at £300m for the biosecurity initiative alone. The BBC notes the lab's full role and remit have yet to emerge, and the government has not said where it will be based or when it will open.

Classic.

The Breakdown, Such As It Is

The rest of the package is easier to pin down. The existing AI Security Institute, set up under the previous government in 2023, gets an extra £120m, which suggests Whitehall sees it as permanent infrastructure rather than a proof-of-concept. That's arguably the most sensible line in the whole announcement.

Cyber defence gets £130m across two new initiatives: the National Cyber Shield and the National Exposure and Real-Time Vulnerability Centre. Both focus on identifying vulnerabilities across critical infrastructure and the wider economy. For UK security firms and managed service providers, this matters: national programmes focused on vulnerability exposure create procurement demand, usually quite quickly.

There's a £60m robotics and manufacturing AI fund exclusively aimed at British companies. A £60m AI research and procurement pot. An £18m chip design academy, which the British Industry group called "a step towards bolstering capability and resilience." A £15m prize for AI solutions to public policy problems, starting with healthcare. Regional mayors are getting £30m to support local AI work, with the National Cyber Security Centre receiving unspecified additional funding to expand in Manchester.

Manchester. Of course. We like that.

The Right Hand Doesn't Know What the Left Is Spending

Here's the bit that made me spit out my tea. In the same week Healey was announcing £600m in AI investment, the Science and Technology Committee published its response, revealing that the government's own finance systems cannot produce consistent annual figures showing how much departments and public bodies spend on digital transformation.

The UK public sector spends more than £26 billion annually on digital technology. The government agreed with the "intent to publish annual data on digital and data-driven spending" while saying it did not "believe this is currently practicable." Existing finance systems could not "provide consistent, reliable and comparable data across departments."

Committee chair Dame Chi Onwurah didn't mince it: at a time when ministers are asking taxpayers to fund significant investment in technology, it's concerning that they cannot show where the money is going or how value for money is being measured.

I spent five years in identity verification and KYC. The whole discipline is built on the premise that you can't manage what you don't measure. You'd think a government announcing a biosecurity lab might have worked that one out by now.

The committee's June report also raised what it politely called "serious concerns" about the growing prominence of Palantir across public services. Quietly, that's the most explosive sentence in the whole parliamentary week and it's getting approximately zero headlines.

The Startup Layer

Meanwhile, at the scrappier, more honest end of the market, things are moving fast. The UK startup funding week ending 23 October saw £173.6m raised across 15 companies, with investors clearly tilting towards agentic AI tied to measurable outcomes. Opply raised £5.5m from Chalfen Ventures for an agentic operating system handling sourcing, procurement, payments and compliance for consumer brands. Azoma secured £8m in a Series A to grow its Agentic Commerce Optimisation platform. Cognitive Credit took £29.87m in growth funding from Summit Partners.

Operators want systems that execute tasks in procurement, commerce and credit analysis, not just produce insights. That's the direction the investable market has decided on. The government would do well to read the fund reports before writing the policy papers.

The Trademark Signal

There's a quieter data point buried in this week's numbers that deserves a mention. AIBD's analysis of IPO trademark data shows Class 42 filings, covering software as a service, technology services and the wider SaaS and AI-tool category, came in at just 2,379 in Q4 2026, down 70.9% versus the prior period. That is a dramatic contraction in brand-new software product registrations.

Two readings are possible. The optimistic one: consolidation is healthy, the chaff is being separated from the wheat. The pessimistic one: founders aren't protecting IP because they're not confident enough in their products, their funding runway, or the market, to file. When I was building SaaS tools in the mid-2000s, the first thing any half-decent lawyer told you was: file the trademark before you file the pitch deck. If that discipline is collapsing, something's wrong upstream.

Given the government just announced a £15m prize for AI solutions to public problems and a £60m procurement pot, there's a potential mismatch brewing: Whitehall is waving money at AI companies while the trademark filing data suggests the formation of new software ventures is drying up sharply.

The Founder's Verdict

I built something like this in 2007. Not the biosecurity lab, obviously. I built a mobile marketing platform for the NHS, which cost me about sixty grand, an ulcer, and eighteen months of procurement meetings. It took two years to get a contract. The system for tracking whether the investment worked? Excel. A shared one, with inconsistent column headers.

The £600m package isn't nothing. A chip design academy is long overdue. The AI Security Institute getting a proper budget is good. The Manchester cyber expansion is welcome, and not just because I'm from the North. But you're in trouble if the same week you announce £600m in new AI investment, you also have to admit you've no idea where the existing £26bn is going.

Opposition critics called the package "a shopping list of projects" without any clear strategy. That's a cheap shot. But it's not entirely wrong. The delivery question is the one nobody at the pre-Budget briefings is answering.

AI policy that hasn't been written by someone who's actually shipped software is Napster without the music. The technology works brilliantly on the slide. The roadblock is always the plumbing.

Ask the government's finance systems.