Britain Prices Its AI Security State

Key points

• HM Treasury’s release of 24 October 2026 itemises £120m more for the AI Security Institute, £60m for a National Cyber Shield, £70m for a National Exposure and Real-Time Vulnerability Centre and £60m for a Sovereign AI R&D Procurement Scheme, all trailed for confirmation at the Budget on 28 October 2026.
• House of Commons Library briefing CBP-10003, published 14 October 2026, records that the UK has no AI-specific legislation and no single AI regulator, and that the AI Security Institute has not been given statutory powers.
• The only AI-specific statutory power in force is section 216A of the Online Safety Act 2023, inserted by section 248 of the Crime and Policing Act 2026; it is unused, and section 249 requires a progress report to Parliament by 31 December 2026.

Britain has put a price on its AI security apparatus, and it is a procurement price rather than a legislative one. On 24 October 2026 HM Treasury published Chancellor unveils Britain’s ambitious plan for AI, itemising a package that John Healey, the Chancellor of the Exchequer, is to confirm at the Budget. Parliament has listed that statement for Wednesday 28 October 2026 at about 12.30pm, after Prime Minister’s Questions. The security, procurement and capability elements covered here amount to roughly £456m in one-off commitments, plus £10m a year to study how AI will reshape jobs and industry. Reuters, reporting on the day, described the wider announcement as “around £600 million ($800 million)”, grouping the two cyber lines into one £130m figure; City A.M. put it at “more than £600m”.

What the money does not buy is a statute. Ten days earlier the House of Commons Library published AI regulation in the UK (CBP-10003), by Elizabeth Rough, which states that “The UK does not have any AI-specific regulations or legislation covering AI as a technology” and that there is no single AI regulator. Read the two documents together and the shape of British AI policy is clear enough: a well-funded testing institute without powers, a cyber sandbox without an act behind it, and a procurement scheme that will set standards through contracts. AI & GPP set out the underlying pattern in deployment before law; the Treasury release is that pattern, costed.

What the £456m buys, and who it answers to

The largest line is £120m more for the AI Security Institute, which the Treasury says will fund testing and research so that Britain stays abreast of what AI systems can do and where they fail. Cyber takes £130m across two bodies: a £60m National Cyber Shield for agentic-AI defence, including a £25m challenge fund and scoping work on a regulatory sandbox covering critical national infrastructure, the public sector and private businesses; and a £70m National Exposure and Real-Time Vulnerability (NERV) Centre, to use AI tools to find weaknesses across critical infrastructure and the wider economy. A further £60m goes to a Sovereign AI R&D Procurement Scheme aimed at health and care alongside AI security and assurance, £60m more to the Sovereign AI Fund, and £18m to a UK Advanced Chip Design Academy. Smaller lines cover a £15m AI Challenge Prize, £23m over three years for an AI Economics Institute and £30m for local AI partnerships with Mayoral Strategic Authorities. A National Cyber Security Centre (NCSC) AI Cyber Security Centre of Excellence in Manchester, and chip design capability in Edinburgh, are announced without figures. There is no data-centre or AI Growth Zone item in the package.

Set each of those against its legal basis and the picture sharpens. The table below is built from the Treasury release, CBP-10003, the Crime and Policing Act 2026 and the parliamentary record for the Cyber Security and Resilience Bill.

Only the last two rows contain law, and neither is in force as an AI rule today.

Why a supplier’s real obligation is a contract clause

For a company selling AI into a British hospital or a transmission operator, the binding instrument in the Treasury package is the procurement scheme, not the testing institute. The AI Security Institute has more than a hundred technical experts and pre-release access to frontier models, as the Minister for Artificial Intelligence, Kanishka Narayan, described in a speech on building a safer, more secure Britain in the age of AI on 12 October 2026. But the Commons Library briefing is blunt about its standing: “To date, the AISI has not been given statutory powers.” An evaluation it publishes is evidence, not an order.

The £60m procurement scheme is different. Money moving through contracts carries conditions, and those conditions are enforceable against the supplier even where no regulator can act. That is how Britain is likely to arrive at de facto AI standards for health, assurance and security: written into schedules, not laid before Parliament. The route is not peculiar to Britain, and AI & GPP has traced how incident-notification duties now reach British operators through sectoral rules rather than any AI act.

Money moving through contracts carries conditions. A testing institute without powers does not.

The government does not dispute the direction. Narayan told his audience: “We will legislate wherever necessary. And we won’t look to law alone.” Healey, in the Treasury release, said he understood and shared the concerns of working people and was setting out what he called a distinctively British approach to AI. Supporters of this model argue that institutional capability, model access and technical testing move faster than primary legislation and are harder for a frontier developer to game. Critics point to the same CBP-10003 ledger: the AI Opportunities Action Plan reports 38 of 50 recommendations met, 76 per cent, while the National Data Library, on a July 2025 ministerial answer the briefing quotes, “remains under policy development” and the copyright consultation’s 11,500 submissions have yet to produce a published response. Britain is also being told it is buying in the wrong register. The Council for Science and Technology, in advice published on GOV.UK on 19 October 2026 as AI sovereignty: preserving UK agency in an AI-enabled world, warns that without deliberate and urgent action the country “risks becoming a high-value AI (artificial intelligence) customer”. That is a sharper version of the question Europe faced when it chose capacity over rulemaking.

The one statutory AI power, and the date attached to it

There is exactly one AI-specific power on the statute book, and it has not been used. Section 248 of the Crime and Policing Act 2026 (c. 20) inserts a new section 216A into the Online Safety Act 2023, letting the Secretary of State make regulations amending that Act to minimise or mitigate risks of harm from illegal AI-generated content and from AI services used to commit or facilitate priority offences. The regulations may impose duties on providers of AI services and give Ofcom functions over them, including services provided from outside the UK. The Act defines an AI service by reference to whether the service, or part of it, can generate AI-generated content.

The clock on it is explicit. Section 249 requires the Secretary of State, “no later than 31 December 2026”, to lay before Parliament a report on progress towards making those regulations — unless a draft statutory instrument has been laid before that date. Nothing in the Treasury package changes that duty, and nothing in the package depends on it. Just over two months after the Budget, the government must either show a draft or explain its absence.

Where the cyber bill actually stands

The Cyber Security and Resilience (Network and Information Systems) Bill is the live legislative vehicle in this area: Narayan’s speech named it as the opportunity to improve the identification and reporting of serious incidents and near misses, and to cut AI-driven risk in essential services. Its position needs stating precisely. Parliament’s bill page, last updated 22 October 2026, shows the current print as HL Bill 49, as amended in Grand Committee, dated 7 September 2026, with Lords committee stage having run from 1 September. The bill has not received Royal Assent. Its Lords Report stage is listed for a sitting on 26 October 2026, with no publications yet attached. In other words the bill is at the door of Report stage, not through it, and the agentic-AI sandbox the Treasury is scoping has no statutory anchor in it or anywhere else.

Three dates decide whether this is a security state with rules or a security state with budgets. The Budget on 28 October 2026 confirms or trims the figures. Lords Report stage, beginning 26 October 2026, determines what incident duties operators of essential services will actually carry. And 31 December 2026 is when the government must account to Parliament for the single AI power it already holds and has not used. Buyers and critical-infrastructure operators should plan on contract terms arriving before any of it.

Questions this answers

What is actually legally binding on a company selling AI to the British state?
Procurement terms, and sectoral rules, not an AI act. The Commons Library briefing CBP-10003 records that the UK has no AI-specific legislation, no single AI regulator and an AI Security Institute without statutory powers, so the £60m Sovereign AI R&D Procurement Scheme sets its conditions through contract rather than statute.

When must the UK government next account to Parliament for AI regulation?
By 31 December 2026. Section 249 of the Crime and Policing Act 2026 requires the Secretary of State to lay a report on progress towards regulations under section 216A of the Online Safety Act 2023 by that date, unless a draft statutory instrument is laid first.

Has the Cyber Security and Resilience Bill become law?
No. Parliament’s bill page, last updated 22 October 2026, shows the current print as HL Bill 49 as amended in Grand Committee of 7 September 2026, with no Royal Assent; Lords Report stage is listed for a sitting on 26 October 2026.

Sources

1. Chancellor unveils Britain’s ambitious plan for AI — HM Treasury, GOV.UK, 24 October 2026
2. UK government sets out $800 million of AI security funding — Reuters, via The Star, 24 October 2026
3. AI regulation in the UK, Research Briefing CBP-10003 — House of Commons Library, 14 October 2026 (PDF)
4. Budget Statement Wednesday 28 October — UK Parliament, 23 October 2026
5. Building a Safer, More Secure Britain in the age of AI — Cabinet Office and Department for Business, Innovation, Science and Trade, GOV.UK, 12 October 2026
6. AI sovereignty: preserving UK agency in an AI-enabled world (web version) — Council for Science and Technology, GOV.UK, 19 October 2026
7. Crime and Policing Act 2026, section 248 — legislation.gov.uk, 2026 c. 20
8. Crime and Policing Act 2026, section 249 — legislation.gov.uk, 2026 c. 20
9. Cyber Security and Resilience (Network and Information Systems) Bill — UK Parliament, last updated 22 October 2026
10. Cyber Security and Resilience Bill: Lords Report stage — UK Parliament, sitting 26 October 2026
11. Healey vows AI will not ‘leave workers behind’ in £600m Budget package — City A.M., 24 October 2026