Shape the stack, or keep renting it
There are two ways for a country to live with frontier AI. Rent it by the API call, or try to shape what gets built and how it is used. The Budget package now being trailed is an attempt at the second. It is smaller, patchier, and more tied to American firms than the slogan suggests.
Ministers have promised a “distinctively British” approach that takes the concerns of working people seriously. No single official total has been published. If someone hands you one neat number for the whole thing, they did the addition. The Treasury has not.
Scotland’s The National has the itemised account. Treat those figures as reported, not as a scored set of appropriations, and do not assume every pound is new money rather than money moved from somewhere else.
What the reported lines actually buy
The clearest cyber item is Nerv, the National Exposure and Real-Time Vulnerability centre, at £70 million. The plan is to use AI tools to spot threats across critical infrastructure and the wider economy. For a board, that is a bet on earlier notice. Earlier notice still depends on someone acting when the alert fires.
A recent outage shows why that distinction matters. The National reports that a PHP security flaw caused disruption across the web, and that 32 UK government services were temporarily offline while the vulnerabilities were addressed. It also reports that the National Cyber Security Centre worked with OpenAI, and that OpenAI’s technology was used to discover the issues and to help contain the fallout.
That is not a story about AI creating the hole. Nothing in the account supports that. Finding a flaw is not the same as restoring the services, and helping to contain the fallout is not the same as owning the fix. The narrower point should bother a buyer: on the defensive side of a British government incident, the tools in the room were American.
The AI Security Institute is reported to be in line for a further £120 million. A further £60 million is reported for the sovereign AI fund, aimed at promising British firms. Another £60 million would back public-service contracts that use the technology, including in the NHS. That last pot is small beside the health budget. A public buyer with a contract is still worth more to a supplier than another strategy paper.
Separately, £10 million a year is reported for work on how AI will change industry, and how Britain should prepare, against explicit fears of job losses. It will not soften a redundancy round. Done properly, it might show which sectors are absorbing the technology and which are cutting headcount and calling it innovation.
Leave a much larger figure out of this. The Chancellor is also expected to promise an extra £20.4 billion for the British Business Bank and the National Wealth Fund, a number first reported by the Financial Times. That is industrial finance. The National does not describe it as AI spending.
A lab announced before its rules
The unusual line is a biosecurity lab officials are calling a “world-first.” The stated purpose is to use AI to help develop medical countermeasures for new pathogens, and to protect against potential future pandemics. Combined backing is up to £300 million: up to £200 million from OpenAI and Anthropic, and £100 million from government.
The split between the two companies has not been set out. Neither have the rules. Who would see biological data, who could stop a project, and what “countermeasures” means in practice are not in the public account. Until they are, this is a political commitment, not a scientific capability.
The National also reports a government view that safeguards have to keep human beings in charge, alongside critics who want a slowdown and unnamed experts warning that, without controls, the technology could wipe out humanity within a decade. No paper, no named scientist, no method. Hold that as a reported fear, not as a finding.
American tools, British letterhead
Put the outage and the lab next to each other. The package gets harder to sell as independence. The tools used when government services went dark were American. A large share of the money for the institution meant to manage biological risk would come from OpenAI and Anthropic. The National describes frontier development as largely led by US and Chinese firms, and frames the spending as a way to depend on that lead a little less. Depending less is not the same as holding the models.
Kanishka Narayan, identified as the AI minister, put the political choice more sharply than the spending lines do. On BBC Radio 4’s Today he said the divide is not pro-AI against anti-AI. It is whether a country is “passive in the face of AI” or “actively shaping it.” He rejected a left that would refuse to build and leave the field to other countries, and a right that would let “tech barons” “let rip” on the economy and society.
That is a useful frame for a leadership team, even if you never vote on it. If core workflows sit on a model you do not govern, trained by a firm whose incentives are not yours, you have already chosen. “Actively shaping it, ” in this package, means British institutions, public contracts and a security centre, still working with American models, and in the lab’s case with American capital. A seat at that table is worth having. It is not the keys.
The government says a wider set of measures, not these AI lines alone, is expected to create more than 100, 000 jobs from 2028-29, including in AI. Read that as a projection. Nobody has hired those people.
The welcome, and the useful complaint
Business groups welcomed the measures. Opposition critics called them a “shopping list of projects” without a clear strategy. A priced list beats a vision with no prices. It is still not an operating plan.
Kate Shoesmith, director of policy at the British Chambers of Commerce, said more sovereign-fund money should strengthen homegrown talent, and pointed to a new UK Advanced Chip Design Academy as a step toward capability and resilience. No location, cost, operator or opening date has been published for that academy. Her more useful point was the next one: government now has to deal with the barriers that stop firms adopting the tools they are being told to celebrate.
Jordan Cummins, a CBI director, made the adoption point from the other side. Firms move faster when they are confident the tools are safe, secure, and working in the workforce’s interests. Plenty are freezing adoption because they do not trust the tools. That is a risk judgment, not a failure of enthusiasm.
Ben Spencer, the shadow AI minister, pushed the argument many executives make in private: AI is built by business, not by government. He wants burdens cut, protectionism rejected, cyber and biosecurity backed, and the AI Security Institute kept in a world-leading position with advance access to models. Advance access is a relationship with the labs that build those models. A £60 million fund does not create it.
Questions for the next board pack
-
How much is the government actually spending on this AI package?
No official total has been published. As reported, the one-off government lines are £120 million more for the AI Security Institute, £100 million for the biosecurity lab, £70 million for Nerv, £60 million for the sovereign AI fund and £60 million for public-service contracts. Those five add up to £410 million. That sum is arithmetic, not a Treasury headline. The lab’s £100 million may sit in a different budget from the cyber lines, and the reporting does not say which figures are new and which are reprofiled. The £10 million a year on labour-market impacts sits on top. The private lab money is separate.
-
Who pays for the biosecurity lab?
Up to £200 million from OpenAI and Anthropic, and £100 million from government, within combined backing of up to £300 million, as reported. The split between the two companies, and the lab’s rules, have not been set out.
-
Did AI cause the PHP disruption?
No. The National reports that a PHP flaw caused global disruption and took 32 government services temporarily offline. OpenAI’s technology was used to discover the issues and to help contain the fallout, with the National Cyber Security Centre involved. Detection is not causation, and help with the fallout is not the same as restoring the services.
-
Is the £20.4 billion part of the AI plan?
No. That expected extra funding for the British Business Bank and the National Wealth Fund, first reported by the Financial Times, is separate. Counting it as AI investment would flatter the technology budget and muddle the capital story.
-
Will this create jobs, or only study the ones AI removes?
The figure of more than 100, 000 jobs from 2028-29, including in AI, is a government projection for a wider programme, not a result of these lines. The £10 million a year is for understanding how work changes. Neither tells you what happens in your own cost base.
-
Which of these pots can my firm bid for, and when?
The trail has not created a route to market. There is no published timetable, eligibility or bidding process for the sovereign fund, the NHS-related contracts, Nerv or the lab. Watch for contract notices, not the speech. If you are buying models for core work, ask who is accountable when the model is wrong. Skills, process and trust will decide more of this year’s results than an academy that does not yet have an opening date.
A Budget can put British institutions a little closer to the work, and it can retire the laziest argument in the room: ban the tools, or hand the keys to the people who built them. It cannot make a company competent. That part is still a management problem. Inconvenient, and the only part you fully control.
Figures that sit outside the package
Public compute and private capital returns are easy to fold into an AI headline. These two reports keep them separate.