The Budget was the main UK AI news this week, and the reports give two different headline figures for its AI money. Technology Magazine headlined a £700m fund for sovereign AI (opens in a new tab) on 28 October, Budget day. Open Access Government headlined a £600m AI investment plan (opens in a new tab) set out by the chancellor and framed around innovation and security, but its own text dates that plan to 24 October, before the Budget and before this week began.
The two figures are not rival counts. The Treasury's Budget document, published on 28 October, itemises AI lines including an additional £120m for the AI Security Institute, £165m for a new National AI Biosecurity Initiative, £154m for cyber defences, £40m for a new National Agentic Security Unit and an additional £140m for the Sovereign AI Unit. City AM reported that much of the more than £700m package was trailed before Budget day, and techUK said the "£600m+ AI package, announced over the weekend, has grown to £700m+". The plan announced on 24 October is part of the total, not a separate pot.
My view is that how the total was built is the story. A UK supplier, public body or transformation team needs to know which part of a headline figure is new money and which was announced earlier. Elsewhere this week NatWest's third-quarter results showed headcount down by about 1,000 full-time equivalent roles in the quarter, A&O Shearman published a note on the ICO's agentic AI consultation, and the EU AI Act's high-risk dates stand as they were set in law in July.
- Term: Sovereign AI
- Definition: Sovereign AI is computing capacity, models and data that a country can run and control under its own jurisdiction, without undue reliance on foreign suppliers.
How much AI money is in the UK Budget 2026?
As of November 2026, the published headlines give two figures: £700m, headlined by Technology Magazine as a fund for sovereign AI, though its text describes an AI package, and £600m for an AI investment plan focused on innovation and security (Open Access Government). The chancellor announced the £600m-plus plan on 24 October, and the Budget on 28 October put the whole AI package at more than £700m, which includes it. The Treasury's Budget documents are the authoritative record.
| Outlet | Headline figure | How it was framed |
|---|---|---|
| Technology Magazine | £700m | Fund for sovereign AI |
| Open Access Government | £600m | Chancellor's AI investment plan for innovation and security |
The difference matters most to people selling to government. A consultancy pricing a bid, or a public sector digital team deciding whether to wait for central funding, is working from a total. City AM reported that much of the more than £700m package was trailed ahead of the Budget (opens in a new tab), and Open Access Government dates a £600m-plus plan to 24 October, so only part of it was new on Budget day. CMS's summary of the Budget (opens in a new tab) lists the £40m agentic unit, £80m for R&D procurement and a £60m AI Pathfinder Fund as the Budget's own announcements.
Is the £600m new this week?
No. The chancellor announced the £600m-plus package on Saturday 24 October, before this week began, in a Treasury press release, and City AM reported it that day as a £600m package. The Open Access Government piece itself dates the announcement to 24 October, so it reports that earlier commitment, not a new one. That is why I don't add the £600m on top of the £700m.
Is the money enough for sovereign AI?
AI Magazine asked whether the 2026 Budget is enough to deliver sovereign AI (opens in a new tab). That depends on the line items: how much goes on compute, how much on grants and how much is reallocated from existing programmes. The Treasury published its Budget 2026 document (opens in a new tab) on 28 October, and it puts around £500m into security and an additional £140m into Sovereign AI: £60m for the Sovereign AI Fund from 2027-28 and £80m for its R&D procurement scheme. My view is that this is mainly a security package with a sovereign top-up, so it shouldn't be judged as a £700m sovereign fund.
Is AI behind UK bank job cuts?
Share Talk reported that NatWest cut 1,000 jobs in three months (opens in a new tab) and set them beside the bank's faster adoption of AI. The number comes from NatWest's third-quarter results, published on 30 October. PA Media reported about 1,000 full-time roles cut in the quarter and 1,400 so far in 2026, as part of an efficiency drive that produced gross cost savings of about £420m in the first nine months of the year and that PA said was helped by investment in technology including AI tools. PA also said the Evelyn Partners acquisition brought about 2,200 staff into the group. PA and Share Talk both reported chief executive Paul Thwaite saying that AI is changing some roles, but the results statement credits "ongoing transformation" for the net reduction, and PA's report of his remarks does not attribute the cuts to AI.
This complicates the opening point. In the same week that the Budget's AI money was reported at hundreds of millions of pounds, a large UK employer reported about 1,000 fewer full-time roles in a quarter, which Share Talk reported alongside faster AI adoption. Yet NatWest's Q3 2026 results (opens in a new tab) put its headcount up by about 800 for the year to date, because the Evelyn Partners acquisition added about 2,200. Public investment in AI capacity and private cost-cutting are going on together, and I can't tell from NatWest's figures how much of the reduction is down to AI.
What is the ICO asking about agentic AI?
The ICO is asking for evidence on how data protection law applies to agentic AI and how organisations manage the risks. Its agentic AI call for evidence (opens in a new tab) opened on 8 October 2026, before this week, and closes on 20 November 2026. A&O Shearman published a note on it this week and I found no new ICO publication on agentic AI, so this is a reminder, not fresh news.
- Term: Agentic AI
- Definition: Agentic AI is software that plans and carries out multi-step tasks, such as sending emails or updating records, with limited human oversight.
A&O Shearman's note on the consultation (opens in a new tab), dated 28 October, also points to a separate ICO report on generative AI. Nothing in that note changes what the ICO is asking.
When do EU AI Act high-risk rules apply?
The EU AI Act's high-risk obligations have already been pushed back in law. The Digital Omnibus on AI, Regulation (EU) 2026/1744, was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026. It moves obligations for stand-alone high-risk systems listed in Annex III to 2 December 2027 and those for high-risk AI in products covered by Annex I to 2 August 2028. These high-risk obligations were deferred, not cancelled, and not everything moved: the Article 50 transparency duties began on 2 August 2026, although generative AI systems already on the market before that date have until 2 December 2026 to meet the Article 50(2) machine-readable marking duty. The European Commission's AI Act page (opens in a new tab) sets out the revised dates.
What the week's UK AI news adds up to
The week's UK AI news was mostly about money and jobs, and both need reading past their headlines. The Budget's AI total includes a package announced before the week began. The NatWest figures come from the bank's own results: Share Talk reported them alongside faster AI adoption, while PA Media described an efficiency drive helped by investment in technology including AI tools. The EU's high-risk timetable is settled in law, not in doubt. A board paper this week would rest on firmer ground citing the Treasury's own Budget document than any of these headlines.
What else launched in AI this week?
Four announcements from 26 October to 1 November 2026 are listed here, each matched to the company's own announcement and to separate reporting. This is not a complete record of the week's launches.
- OpenAI introduced GPT-6.1 Terra on 26 October 2026, a medium-sized model it pitches on fast computer use at a lower price. GitHub made it generally available in Copilot the same day, and 9to5Mac reported it was available the same day through ChatGPT Work, Codex and the API.
- Google announced Gemini 4 Carbon on 28 October 2026, which its launch post (opens in a new tab) calls "our next frontier model". Access is limited at first: Constellation Research reported it will go to cyber defenders before wider access through Google's Fairwind Program, and VentureBeat reported an introductory price of $2 per million input tokens and $10 per million output tokens.
- Google DeepMind announced Genie 4 on 27 October 2026, a real-time world model that generates audio natively. It is a research preview for trusted testers.
- Microsoft AI released MAI Image 2.7 in preview on 27 October 2026, an image editing model, with text-to-image generation to follow.