UK AI package: £120 million for the AI Security Institute and where the other £500 million goes

HM Treasury published a roughly £600 million AI package on October 24, 2026, and the biggest single message in it is security, not growth: £415 million of the itemized total goes on AI security, including £120 million more for the AI Security Institute (AISI). The Treasury page does not give one headline sum, but its itemized lines add up to about £621 million, which matches the "£600m+" description from techUK and the roughly $800 million reported by Investing.com.

For a package this size, the detail is where the information is. Here is the breakdown, what is new, and what is missing.

The numbers, line by line

The Treasury page splits the money into three groups.

AI security, £415 million. This is £165 million for the National AI Biosecurity Initiative (of which £100 million is the new AI Biosecurity Lab and £65 million is surveillance and countermeasures for priority pathogens); £120 million more for the AI Security Institute for testing, research and technical standards; £60 million for a National Cyber Shield that uses agentic AI for cyber defense, including a £25 million challenge fund and scoping for a regulatory sandbox for critical national infrastructure; and £70 million for a National Exposure and Real-Time Vulnerability Centre that uses AI to find weaknesses in critical infrastructure.

Homegrown capability, £138 million. An extra £60 million for the Sovereign AI Fund, with a new focus on physical and industrial AI such as robotics and advanced manufacturing; an extra £60 million for a Sovereign AI R&D Procurement Scheme covering health, care, AI security and assurance; and £18 million for a UK Advanced Chip Design Academy.

Good growth in every postcode, £68 million plus unpriced items. A £15 million AI Challenge Prize whose first award targets health; an extra £23 million over three years for an AI Economics Institute; £30 million for local AI partnerships with mayoral authorities; and a National Cyber Security Centre AI centre of excellence in Manchester with no figure given.

What is notable

The AI Security Institute gets a clear top-up. At £120 million it is the second-largest line, behind the biosecurity initiative. Our explainer on how the UK and US testing institutes differ covers what AISI tests. The money matters less than the access: AISI's recent work has depended on getting models before release, and on running evaluations safely, a theme in our report on its cyber evaluations.

There is no compute announcement. The page gives no capacity figures. The only compute-linked commitment is up to £200 million in model credits from OpenAI and Anthropic for the biosecurity lab, which is a company pledge, not government infrastructure. That makes this a package about security and adoption, not about building a national compute base. For the earlier record of what has been promised versus delivered, see our funded versus announced analysis.

Public-service results are used as the justification. The page cites AI analysis of more than 210,000 NHS chest X-rays a month, about a third of the total, with suspected lung cancer identified within 24 hours up 11.5 percent, roughly 400 more patients a month. It also cites about 6.3 million customer interactions on HMRC's Ask HMRC assistant in 2025-26. These are government-reported figures, and the page does not link the underlying evaluations.

What the officials and industry said

Chancellor John Healey says "AI will transform daily life: every firm, community, hospital, school, and every aspect of our economy can benefit." AI Minister Kanishka Narayan says "We cannot afford to rely entirely on technology built overseas." Security Minister Dan Jarvis frames the security lines as "staying one step ahead of those who seek to use AI to do us harm." techUK chief executive Julian David called it "a strong £600m+ package focused on AI adoption, trust and security, investment and AI's impact on jobs and society."

Caveats worth knowing

The page says the Chancellor "is set to" announce the package "next week" while carrying a date of October 24, and says it forms part of his Budget. The timing is unclear, so lines could be re-announced or adjusted when the Budget lands. Several items, including the Manchester centre of excellence, carry no figure. And "additional" means top-up: the page does not say what the earlier totals were for the Sovereign AI Fund or the Institute.

The arithmetic

The Treasury page does not total itself, so here is the sum. AI security: £165 million for biosecurity, £120 million for the Institute, £60 million for the Cyber Shield and £70 million for the vulnerability centre make £415 million. Homegrown capability: £60 million plus £60 million plus £18 million make £138 million. Good growth in every postcode: £15 million plus £23 million plus £30 million make £68 million. Together that is £621 million, before the unpriced Manchester items and before the £10 million a year for work on jobs.

Investing.com groups the lines differently, putting the Cyber Shield and the vulnerability centre together at £130 million, and calls the whole about £600 million. The two descriptions agree; the difference is rounding and grouping. The company credits are on top of this: up to £200 million from OpenAI and Anthropic is a pledge, not Treasury spending.

What to watch

Watch the Budget for these lines to reappear with dates. Watch whether the Institute's top-up is paired with a published access arrangement for frontier models. And watch whether the unpriced items, especially the Manchester centre of excellence, get a figure.

Our take

This is a security-first package with a real number for the institute that tests the labs' models, and that is the most useful line in it. The weakness is what it leaves out: no compute plan, no access rules, and a timing note that makes the whole thing look provisional. Judge it in the Budget, when the lines either appear again with dates or do not.