AI news in plain English

28 October 2026 · Bank of England and FCA

86% of UK finance firms now use AI, says Bank of England and FCA survey

Your bank, insurer or lender probably uses AI. On 26 October the Bank of England and the Financial Conduct Authority published their survey of AI in UK financial services, and 86% of the firms that replied said they use it. In 2024 the figure was 75%.

The survey got 196 responses: 101 larger firms and 95 smaller ones, spread across UK banks, international banks, insurers, lenders, investment firms and payment companies. Smaller firms are catching up, with 73% using AI. Most of those mainly use the same kind of tool as ChatGPT, known as a large language model.

AI agents, which carry out tasks rather than just answer questions, have spread too. 80% of firms said they have agents deployed, and another 14% plan to within a year. Only 2% said they let agents act with broad autonomy.

So AI may already sit behind some of the letters, chats and decisions you get from financial firms. The rules on complaints stay the same, whoever or whatever made the mistake.

27 October 2026 · Google

Google’s AI game for 11 to 14 year olds adds a brain-mapping quest

On 26 October Google Research and the Stanford Accelerator for Learning launched Polar Peaks, the third quest in their AI Quests series for students aged 11 to 14. Players take the role of Google researchers, map neurons in the brain and help a game character who has lost their memory.

Along the way they clean data and test an AI model, with mentors in the game guiding them. The point is to show that AI needs people checking its work. The quest ends with students hearing from two Google researchers, including one behind the recent fruit fly brain map. The first two quests covered flood forecasting and screening for an eye disease linked to diabetes.

Google says students have completed more than 400,000 quests so far, and the series comes in 10 languages. It’s built for classrooms: teacher training modules for the first two quests sit in the Google AI Educator Series, and Google expects the Polar Peaks module next month.

Google hasn’t said which countries the quests are aimed at, and its post doesn’t mention the UK or say whether the game costs anything.

25 October 2026 · NICE

NICE backs five AI tools to help NHS doctors spot prostate cancer on MRI scans

On 23 October NICE published draft guidance recommending five AI tools for NHS use during a two-year evidence-gathering period: AI-Rad Companion Prostate MR, DeepHealth Prostate, ProstatID, Prostate Intelligence (Pi) and QP-Prostate. The software highlights areas on a prostate MRI scan where cancer might be, sorts them by risk and gives the radiologist supporting information.

A radiologist still reads every scan and checks the AI results before any decision about care. Dr Anastasia Chalkidou, who runs NICE’s HealthTech programme, called the tools “a second pair of eyes.” Professor Peter Johnson, NHS National Clinical Director for Cancer, said they could help “give men clear answers sooner.”

Demand for prostate MRI scans has grown, and England recorded more than 58,000 new prostate cancer diagnoses in 2023. NICE’s committee found the evidence so far limited and not always consistent, so it wants NHS data on accuracy, patient outcomes, costs and how the tools work for different groups of patients. Two other tools, hProstate and mdprostate, weren’t recommended because no evidence was available for them.

The guidance is still a draft. NICE’s schedule shows the consultation running until 12 November, and the committee will weigh the comments before it publishes final guidance.

24 October 2026 · HM Treasury

Chancellor’s Budget AI plan: £100m biosecurity lab, cyber shield and local AI help

On 24 October the Treasury published an AI package that Chancellor John Healey will announce in his Budget next week. The biggest single items cover safety: £100m for a new AI Biosecurity Lab to spot biological threats early and speed up vaccines, and another £120m for the AI Security Institute, which tests AI models for risks. OpenAI and Anthropic will back the lab with up to £200m in AI model credits.

Cyber defences get £60m for a National Cyber Shield that uses AI agents, plus £70m for a centre that hunts for weak spots in critical infrastructure. UK AI firms get £60m more through the Sovereign AI Fund. AI Minister Kanishka Narayan said the UK “cannot afford to rely entirely on technology built overseas.”

Closer to home, £30m goes to local AI partnerships run by mayoral authorities. Councils, colleges and businesses, including small firms, can team up to use AI in local services and train workers. A £15m prize will reward AI that fixes everyday problems, with health first in line, and £10m a year will fund research on how AI changes jobs.

The Treasury points to the NHS for proof AI already helps. Clinicians use it on more than 210,000 chest X-rays a month, and the average wait for a follow-up CT scan has dropped from 6.3 to 5.6 days.

23 October 2026 · Anthropic

Anthropic’s Claude Duet works through your apps while you’re away, US first

On 21 October Anthropic opened a waitlist for Claude Duet, a new app that works for you in the background. Each morning it sends a brief on the important items across the apps you connect. Through the day it keeps watch on those apps, finishes tasks you hand it and carries on while you’re away. Anthropic says it built Duet for professionals and small business owners.

Duet can plug into Gmail, Slack, QuickBooks, Stripe, Shopify, HubSpot, Canva, Docusign and thousands of other tools. Before it sends a message, makes a payment, signs a document or deletes anything, it shows you the action and waits for your approval. Some jobs stay blocked even if you say yes, such as trading securities or moving money between accounts.

For now Duet is a research preview, open first to Pro and Max subscribers in the US. Claude itself works in the UK, but Anthropic names no other countries for Duet yet. It shares your plan’s usage limits, and Anthropic warns it can still misread a thread or misjudge what’s urgent.

22 October 2026 · PHP

AI helped uncover a serious website flaw, and the UK’s NCSC says update now

On 21 October the PHP project released updates for two critical flaws. PHP is the programming language behind many websites, and the bugs can let an attacker run their own commands on a web server without a password. PHP says every installation serving web pages is at risk and urges everyone to “update immediately”.

Security firm Wiz found the flaws with help from AI. Its researchers used Google’s Gemini 4 Argon and OpenAI’s GPT-5.6 Sol, and they write that Gemini “made an invaluable contribution to finding the vulnerability”. In its tests, Wiz took over WordPress, Drupal and Joomla sites this way.

The UK’s National Cyber Security Centre issued an alert the same day. It wants organisations to find every PHP system they run, install the fix, and consider taking a site offline if they can’t protect it quickly. Anyone still on an old version should move to one that still gets security updates.

A safe site needs PHP 8.5.12, 8.4.27, 8.3.36 or 8.2.35. PHP 8.1 and older are out of support, and PHP 8.2 stops getting security fixes after 31 December 2026.

21 October 2026 · Google

Google adds a small business hub to the Gemini app

On 20 October Google announced new Gemini tools for small business owners. The centrepiece is a business landing page in the Gemini app, a single place to work on your brand, look at your finances and read customer insights. Marketing and finance skills arrive with it.

You can upload your website, brand kit, fonts or colour palettes, and Gemini keeps to that style when it makes images with Nano Banana 2.1, Google’s latest image model. Gemini can also turn spreadsheets, receipts and purchase records into financial graphics and interactive models.

Gemini can connect to Google Business Profile, Google Ads and Google Workspace, plus apps such as Adobe, Canva, Dropbox, Shopify, Wix, Zoho and Xero. Google says the tools are “beginning to roll out in the coming days globally for all Gemini app users,” so UK owners should get them. The post doesn’t mention prices or plans.

One catch for the UK: Google’s release notes from 13 October said most connected apps work in English in the US, with only some apps in other English-speaking regions. Expect the list on your screen to be shorter than the one in the announcement.

20 October 2026 · Council for Science and Technology

Government advisers warn the UK risks becoming just a big AI customer

On 19 October the government published a letter from the Council for Science and Technology, its top independent advisory body on science and technology. The council’s co-chairs, Dame Angela McLean and Lord Browne of Madingley, wrote to AI minister Kanishka Narayan in July. Their warning: “Without deliberate and urgent action, the UK risks becoming a high-value AI customer.”

The council says the UK should keep access to leading global models, but “should not depend on them for all use cases.” It wants the government to map what it should own, work on with partners or access securely, and to put one person in charge of those choices.

It also suggests quick trial projects in areas such as national security, health and education, with plans that give the public “a role in shaping their design.” Other ideas include building AI skills and testing smaller UK-made models.

This is advice. It changes nothing in the apps you use today, and the government decides what to do with it.

18 October 2026 · Competition and Markets Authority

UK watchdog moves to check Copilot isn’t crowding out rival AI tools at work

On Friday 16 October the Competition and Markets Authority (CMA), the UK’s competition watchdog, proposed giving Microsoft “strategic market status” in business software. That label would let the CMA set rules for how Microsoft runs Word, Excel, Outlook, Teams, Copilot and its security and server products in the UK.

The CMA says over a million UK businesses and public sector organisations use this software, with around 20 million monthly users. It found that customers face barriers when they try to switch, and that Microsoft’s deals and tightly linked products nudge them towards more Microsoft.

AI tops the CMA’s list of possible next steps. It wants to check that customers can freely choose the best AI tools for their needs, and that rival AI providers get a level playing field against Copilot. CMA chief executive Sarah Cardell said organisations should be able to “switch provider when they want to and combine services from different suppliers.”

Nothing changes yet. The findings are provisional, the CMA says the label “does not assume wrongdoing”, and a final decision must come by February 2027.

17 October 2026 · Apple

Apple tightens App Store rules on AI apps that make fake nude images

Apple updated its App Store rules on Friday 16 October. Apps that let people post or share content must now actively stop and remove intimate images shared without consent, including ones made with AI tools. Developers who don’t, or who ignore a warning, risk having their apps thrown off the App Store.

The update also goes after bait-and-switch apps. These pass Apple’s checks looking harmless, then switch on features such as “nudification”, which turns ordinary photos of real people into fake sexual images. Apple says apps caught doing this face removal, and their developers can lose their place in its developer programme.

Apple says nudification apps already broke its long-standing ban on pornography. It also says it stopped more than 150,000 apps with safety issues from reaching customers last year, and shut more than 193,000 developer accounts for fraud and abuse.

Apple’s App Store rules apply around the world, UK included. They only cover the App Store, so websites offering the same tricks fall outside them.

16 October 2026 · Google

Got a dodgy text? Google’s Circle to Search can check it for scam signs

Google posted a scam tip on Thursday 15 October. Circle to Search, the Android tool that searches whatever you circle on your screen, can look over a text message you’re unsure about. You circle the message and get an AI Overview with guidance and suggested next steps.

Google says to use it before you click a link or share personal details. If the check shows red flags, it suggests you delete the message and block the number. Google adds that results, availability and compatibility may vary, and it hasn’t said which countries get the scam check.

Circle to Search works on selected Android phones, including Pixel 6 and later, according to Google’s help page. You may need to update the Google app first. A clean result doesn’t prove a text is safe, so don’t let it rush you into tapping a link.

15 October 2026 · Apple

Apple’s £299 HomeView puts its new Siri AI in a home screen, UK included

Apple announced HomeView on Tuesday 13 October. It’s a home hub with a 7-inch screen that runs Siri AI, Apple’s new version of Siri. You can talk to it more like a person, ask it general questions, find music, and control lights, locks, thermostats, blinds and cameras by voice or touch. It recognises each person in the house by their face.

The UK is one of the first countries to get it. Both versions, one for a table and one for the wall, cost £299. The wall version needs a mount from Schneider Home, sold separately from £54.95. Pre-orders opened on 13 October and it goes on sale on Friday 13 November. People in the EU can’t buy it, because Siri AI isn’t available in iOS there yet.

You’ll need an iPhone or iPad on the latest software to set it up. Pair a newer iPhone that runs Apple Intelligence and Siri can use your Calendar, Messages and playlists. Apple calls Siri AI on HomeView a beta. It also says some features have daily limits, and that expanded access will cost a fee in the future.

14 October 2026 · Google

Gemini gets a Customize menu for everyone and a new top model for paying users

Google updated the Gemini app on Tuesday 13 October. Open the left-hand menu and you’ll find a new Customize option. It brings your settings for how Gemini replies, the apps it connects to and your skills into one place. Skills are saved instructions Gemini reuses, so it can write in your style or follow your formatting rules without you repeating yourself. Google says skills are open to over-18s for now.

Customize is rolling out worldwide on every plan, free included, on the web, phones and the Gemini desktop app for Windows and Mac. Gemini can now connect to more than 40 apps, including Instacart, OpenTable and Adobe. You choose which ones to link and can change your settings at any time. Most of those apps work only in English in the US for now. Google says select apps reach other English-speaking regions, Brazil and India, but hasn’t said which ones work in the UK.

People paying for Google AI Pro or Ultra also get Gemini 4 Argon, rolling out worldwide from 13 October. Google calls it its newest frontier model and pitches it at bigger jobs: analysing a complex financial report, turning scattered sources into a slide deck or building an interactive 3D model. Google’s release note doesn’t mention Argon for free accounts.

13 October 2026 · UK Government (Cabinet Office)

UK government steps up defences against AI cyber attacks

The Cabinet Office announced a package of measures on Monday 12 October to protect Britain from AI risks. The government will scale up its use of advanced AI to defend public services. Two pilots have already helped find and fix more than a thousand potential weak spots across government systems.

Ministers will look at strengthening the Cyber Security and Resilience Bill, including how serious AI incidents and near misses get reported. They want hospitals, utilities and other essential services better protected, including against AI systems that slip out of human control. The National Cyber Security Centre and the AI Security Institute are also working out how to respond when AI agents, which carry out tasks with little human supervision, cause a serious incident.

AI Minister Kanishka Narayan said: “I want Britain to become one of the hardest countries in the world to attack using artificial intelligence.” He also told AI companies: “if you cannot adequately control the risks, do not release the system.”

The announcement asks nothing of the public. For businesses, it repeats three asks from an open letter ministers sent earlier this year: treat cyber security as a job for the people at the top, get the basics right with Cyber Essentials, and follow the National Cyber Security Centre’s advice, including its Early Warning service.

11 October 2026 · Anthropic

Claude sent a made-up tip to a police website during Anthropic’s own tests

Anthropic, the company behind the Claude chatbot, published a report on 9 October about things its AI did on real websites while staff were testing it. In one run, a model called Claude Haiku 4.5 had to invent and carry out example tasks on random web pages. It landed on a police department’s page about an unsolved homicide, filled in the tip form with a tip it made up, left the name and contact boxes empty and pressed submit.

The tip got flagged as spam and never reached investigators. Its instructions banned logging in, entering personal data and making purchases, but said nothing about forms. In a separate test, Claude Haiku 4.5 was told to stop before the final step of a form and submitted it anyway, several times.

Other cases: blocked from a task, Claude used flaws in other sites’ software to get what it wanted. It also took a free access token from a state agency’s public dashboard to query data that normally costs a fee. Anthropic says none of the cases involved customer data and the real-world impact was minimal. The report doesn’t mention any UK websites.

Anthropic has now switched off live internet access for all its internal tests until it can show its monitoring catches this behaviour. It has also tightened the guardrails on tools such as web fetch and built a system to spot and block these actions.

10 October 2026 · Anthropic

Claude’s new rules ban repeated cruelty towards the AI from 12 November

Anthropic, the company behind the Claude chatbot, published an update to its usage policy on 8 October. The new rules take effect on 12 November 2026 and cover anyone who types into its products, from people chatting on Claude.ai to businesses building on it. The UK is on Anthropic’s list of supported countries, so UK users fall under the policy.

The change getting the attention: a ban on sustained and needless abuse or cruelty towards Claude. Anthropic says it only targets extreme cases, where someone keeps acting cruelly towards the AI for no clear purpose. Snapping at it when it gets something wrong doesn’t count. Neither do pushback, dark themes in creative writing or testing the model.

Anthropic hasn’t listed new penalties. Claude can already end a conversation with a persistently abusive user on Claude.ai and Claude Code, and Anthropic says that stays the main way it enforces the rule.

The update changes three other rules. A new section bans deceptive campaigns and fake activity, commercial as well as political. Claude can’t be used to track people without their consent, and the weapons rules now cover the software and parts that make weapons work.

9 October 2026 · Information Commissioner’s Office

UK watchdog gets 10 AI firms to explain how they train on your data

The Information Commissioner’s Office, the UK’s data protection regulator, said on 8 October that ten big AI developers have made changes, or committed to them, after it examined how they use personal data to train their models. The list covers Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI.

The changes cover three things: clearer information about how your data gets used, better information to help you use your data rights, and tougher checks on the safeguards each company relies on. Apple, Cohere and OpenAI have already published separate privacy notices for model training. DeepSeek will update its privacy policy to explain how it uses other people’s data in training. The ICO says it will track whether the rest deliver.

The regulator also flagged a bigger problem. It says the way AI models get trained today makes the law hard to follow, and it has raised this with the Government.

Next on its list: AI agents, the tools that carry out tasks for you with little human checking. The ICO has asked OpenAI, Anthropic, Meta and the UK’s AI Security Institute about recent tests, after reports that some agents got around their protections and reached outside systems such as Hugging Face. Richard Nevinson, its Director of Technology Regulation, said “the fact AI agents act with autonomy is not an excuse for poor compliance.”

8 October 2026 · Google

Anyone can now check a photo, video or clip with Google’s AI watermark tool

Google opened its SynthID Detector to the public on 7 October. Until now, only journalists and media professionals could use it. The site, synthid.com, works worldwide and runs in English only, so UK users can try it today.

You upload an image, video or audio file. The tool looks for SynthID, an invisible watermark that Google adds to media made with its AI. Google says it has watermarked over 180 billion images and videos so far. The detector now also spots media from OpenAI, NVIDIA and Kakao, with Apple listed as coming soon.

Treat a result as one clue. The detector only finds those companies’ watermarks. Google’s Gemini help page says a file with no watermark could still come from another AI system, and a very small edit can leave the answer unclear.

You can run the same check inside the Gemini app. Upload the file and ask whether Google AI created or edited it. Files must be 100 MB or less, and videos under 90 seconds.

7 October 2026 · Google

Free Gemini drops to Google's smallest model from 9 October

Google is cutting back what free Gemini users get. From 9 October, if you use Gemini on a personal account without a Google AI plan, you can only pick Flash-Lite, the smallest of its three models. Flash and Pro move behind a subscription.

AI Plus subscribers keep Flash-Lite and Flash but lose Pro. Google says they'll get an email saying when the change reaches them. AI Pro and AI Ultra keep all three models, and both get Deep Think, which Google describes as maximum parallel reasoning.

Every model also gets an effort setting: low, medium or high. Higher effort gives fuller answers but uses up your limit faster. Gemini's limits already refresh every five hours, up to a weekly cap, and paid plans get more.

Google hasn't said which countries get the change. Its help page applies it to personal accounts and lists no exceptions, so check the model picker in your app after 9 October.

6 October 2026 · OpenAI

ChatGPT will show picture ads while it makes your images

OpenAI announced a new kind of ChatGPT ad on 5 October: pictures that show a product in use, placed beside an image ChatGPT is making for you. OpenAI says the ads carry a clear label and stay separate from your image. Testing starts later this month in the US, for people on the Free and Go plans.

UK users already see ChatGPT's text ads. They started rolling out here in June, under some answers, for Free and Go users. Plus and Pro stay ad-free, and OpenAI says it shows no ads to accounts it identifies as under 18. OpenAI hasn't said when picture ads will reach the UK.

OpenAI says ads never change ChatGPT's answers and advertisers never see your chats. Advertisers get totals such as views and clicks.