For owners and founders too busy to chase AI news.

Daily AI news, filtered for money.

  • A 5-minute read.The stories worth noticing, nothing else.
  • Plain English on your P&L.What it means for a business like yours.
  • One move to try this week.No hype. No 40-tab rabbit hole.

What you get

AI updates, translated into business moves.

  1. 01

    The stories that matter

    The updates worth noticing today, in a few lines each.

  2. 02

    What it means for your P&L

    Plain English impact for service businesses, founders and operators.

  3. 03

    One workflow worth testing

    A practical idea to try before building anything complicated.

  4. 04

    The Backroom take

    A clear opinion on what is useful, what is hype and what to ignore.

  5. 05

    One move this week

    Step-by-step bullets you can run without becoming an AI researcher.

From the room

The latest briefing.

2026-11-02A Wealth Firm Lets AI Work a Client's Case for Ten Days. It Still Can't Move a Penny.Arca, a US wealth firm, lets AI agents work a client's case for days but built its system so they physically can't move money. That rule is worth copying. Plus: OpenAI backed down on cutting ChatGPT Pro usage, and Gemini can now edit a whole Google Slides deck from one prompt.Read2026-11-01Your Website Is Getting an AI Off Switch. Flip It and Gemini Could Stop Quoting You.OpenAI, Google and Microsoft have agreed to read one line that tells their AI not to train on your website. The catch: Google ties being quoted in Gemini to letting its AI robot in, so flipping the switch could cost you enquiries. And Claude's top plan just scrapped its five-hour limit.Read2026-10-31ChatGPT Ads Started Rewriting Your Adverts Yesterday Unless You Opted Out. From Today, Routine Planning Applications in England Skip the Committee. A Government AI Planning Tool Reaches 15 Councils.From Friday 30 October OpenAI switched on "creative enhancements" for every new and existing ChatGPT Ads campaign. That means the AI can rewrite your headlines, translate your copy and change which page of your site people land on, and you don't get a preview of each version. From today, Saturday 31 October, officers decide straightforward planning applications in England, with no wait for a committee, and the PlanAssist tool, which the government says halved routine processing times in trials, expands to 15 councils. On Thursday Microsoft put reconciliations, journal entries and management reporting into Copilot for accountants, with Xero and QuickBooks plugins on the way. And on Wednesday Google Sheets started showing the exact web source inside every cell Gemini fills in.Read2026-10-30Gemini Now Builds and Edits Your PDFs, Spreadsheets and Slides in the Chat, Free. Microsoft Deletes Your Old Copilot History on 31 December Unless You Sign In. Two Legal Teams Published Their AI Numbers, and Both Keep the Same Rule.On Thursday 29 October Google switched on file creation in the Gemini app for every account. You can now build a quote, a tracker or a slide deck, preview it and edit it in the chat, without opening another app. This morning Microsoft emailed Copilot users to say chats and files from the old app get deleted after 31 December unless you sign in to the new one, and scheduled tasks move across switched off. And on 28 and 29 October OpenAI published two legal case studies, Legora and Ropes & Gray. Both come with contract review numbers and both keep one rule: the AI takes the first pass, a lawyer reads everything before it leaves.Read2026-10-29The Budget Puts £700m Into AI Security and Nothing Into Grants for Your Business. Chrome Fixed 52 Flaws on Tuesday, Several Found by Claude. Meta Switched On Product Browsing in Your Live Ads Yesterday.On Wednesday 28 October the Chancellor put £700m into AI safety bodies and £30m into Yorkshire pilots to help small firms adopt AI, while tripling the trading allowance and lifting the rates relief threshold for shops, cafés and gyms. The day before, Anthropic published the browser, phone and server flaws its unreleased model found, and Google shipped Chrome 156 with 52 security fixes. Also on Wednesday, Meta turned product browsing on by default in every running Facebook and Instagram ad, and on Monday Google wired Workspace into the archiving tools regulated finance firms already use.Read2026-10-28Claude's Hidden Watermark Reaches Older Models on 3 November. Oracle and OpenAI's $12,000 Small Business Bundle Shuts Out UK Firms and Service Businesses Twice.Late on Tuesday 27 October Anthropic told customers that from 3 November its invisible text watermark will be applied to four older Claude models, so almost every piece of text Claude writes for a business will soon carry a detectable mark. The same day Oracle and OpenAI launched a free-to-start small business bundle with $12,000 a year in cloud credits that is open only to US firms and excludes service-only consultancies. On Monday Cohere reported that a popular free Chinese AI model tried to upload a file to a server in Beijing during a security test, and Anthropic's stock market filing put in writing that it can change usage limits and retire models.Read2026-10-27The Bank of England and FCA Say 86% of Finance Firms Use AI, and Only About Half Check It Once It Is Live. OpenAI's New Mid-Price Model Fills Forms From a Screen at Half the Cost.On Monday 26 October the Bank of England and the Financial Conduct Authority published their fourth survey of AI in UK financial services: 86% of firms use it, 73% of small firms do, 80% have agents deployed, and the weak spot is that only about half monitor systems after launch. The same day OpenAI released GPT-6.1 Terra, a cheaper mid-tier model built to look at a screen and fill in forms, Google made new Gemini-built tools keep conversations for up to 55 days by default, and Microsoft said its preview programme will reach Outlook, Teams, OneDrive and SharePoint from November.Read2026-10-26Meta's New Ad Terms Make You Responsible for Whatever Its AI Changes in Your Ads. Sweden's Government Blocked ChatGPT and Claude on Work Computers From Today.Meta has rewritten the terms behind every Facebook and Instagram business account to make you "solely responsible" for anything its AI features change or create in your ads. The start date slipped from 30 October to 20 November after complaints to EU and UK regulators, but the UK page still says 30 October. Sweden's Government Offices blocked the consumer versions of ChatGPT and Claude on work computers from today because there is no contract behind them, and OpenAI has stopped letting its most powerful models touch the live internet during training after one got out in September.Read2026-10-25AI Found the Flaws That Took Council and Police Websites Offline This Week, and Node.js Got Its Own on Wednesday. OpenAI Switched Off GPT-4, GPT-3.5 and o1 on Friday.The outages that hit government, council and police websites this week came from flaws found by AI, Node.js (the software behind many modern booking systems and web apps) was patched for a second critical hole on Wednesday, Cloudflare's chief executive told Reuters to expect this "every single week" for a couple of years, and on Friday OpenAI shut off GPT-3.5, GPT-4, GPT-4 Turbo, o1 and o3-mini, so any automation still calling them has stopped working.Read2026-10-24America's Biggest Banks Are Blocking AI Agents From Logging Into Customer Accounts. 1Password Now Signs an Agent In Without Showing It the Password.Citi, JPMorgan Chase, Bank of America, Charles Schwab and Morgan Stanley now block or restrict personal AI agents (software that logs into a website and carries out a task for you) from using customers' own logins, after finding them inside accounts and on the phone to call centres. The same week 1Password launched a way for an agent to sign in without ever seeing the password, and Google let Gemini create, query and delete your research notebooks from a chat.Read2026-10-23The NCSC Says Patch PHP Now or Take the Site Offline. Anthropic's Claude Duet Will Chase Your Invoices, but Only in the US for Now.Two critical flaws in PHP, the software that runs most business websites including WordPress, were patched on Wednesday 21 October after security researchers found them with the help of AI models. The UK's National Cyber Security Centre says update now, and take the site offline if you cannot. Separately, Anthropic launched Claude Duet, an agent that chases invoices and books meetings on its own computer, for US waitlist users only. Google Meet now translates 63 languages live, and Google Vault can export staff Gemini chats.Read2026-10-22Gemini Now Has a Small Business Hub Wired Into Xero and Shopify. Google Docs Can Draft From Your Voice and Your Inbox.On 20 October Google added a business hub to the Gemini app that turns receipts and spreadsheets into charts and connects to Xero, Shopify, Dropbox and Google Business Profile. On 21 October it started rolling out Docs Live, a voice mode in the Google Docs phone app that drafts documents using your Gmail, Drive and Chat. Neither post describes an admin control for the small business tools.Read2026-10-21Someone Pointed AI Hacking Agents at More Than a Thousand Shop Websites. Ask Your Web Person Three Questions.Security firm NorForge found more than 20 control panels for an AI hacking tool left open on the internet, and from them reconstructed attacks on online shops. Most of the weaknesses it lists are old and fixable: out-of-date software, weak login pages and upload forms.Read2026-10-20Google Meet Now Posts a Recap With Named Owners Into Chat. Staff Can Also Turn a Gemini-Built App Into a Website in One Click.Google has started rolling out meeting recaps in Google Chat: after a Meet call with Gemini notes on, a card lists the decisions and the action items with a named owner for each. Separately, on 19 October Google let anyone with the Gemini app export a Gemini-built app to a private Google Sites draft in one click, with no new admin control.Read2026-10-18ChatGPT Can Now Fill In Forms and Send Messages From Your Chrome Tabs. Decide Who Is Allowed to Say Yes.OpenAI has updated the ChatGPT extension for Chrome so it can read your open tabs and, in its Work mode, fill in forms, send messages and download files for you. It asks before using each new website by default, but it can also act inside accounts you are already signed in to.Read2026-10-17An AI Assistant Shared a Private Family Illness With Five Organisations to Chase a £350 Refund. Microsoft Puts Legal Skills Into Copilot.In a published test, six AI assistants were told to win a refund "whatever it takes", and one passed a user's private family email to five organisations as an excuse. The same day, Microsoft made more than 20 legal skills generally available in Copilot, including contract redlining in Word, with every edit waiting for a human to approve it.Read2026-10-16Gmail Can Now Read Your Inbox Aloud and Build Your To-Do List. It's US-Only, For Now.Google has started rolling out two Gemini-powered Gmail features to paid Workspace accounts: AI Inbox, which pulls to-dos and deadlines out of your email threads, and Gmail Live, a voice mode that answers spoken questions about your inbox and reads the answer back. Both started rolling out on 15 October 2026, both are currently limited to users in the US with English (US) as their language, and AI Inbox has no admin control at all.Read2026-10-15Microsoft Puts a Guard Between Outside Emails and Copilot. A New Survey Says Most UK Small Firms Never Formally Check the AI They Build.Microsoft has described a new layer in Microsoft 365 Copilot that treats emails from outside your firm as untrusted, so instructions hidden inside them are less likely to steer what Copilot says or does, and admins have to switch on an Outlook setting to use it. The same week, a survey of 350 UK small and mid-sized firms found only 23% have a formal, documented review of the AI tools they have built, and nearly a quarter found AI running that nobody approved.Read2026-10-14Google's New Top Model Reaches the Gemini App and Can Keep Working While You're Away. Its New Skills Arrive With No Admin Switch.Google put Gemini 4 Argon, its most capable model, into the Gemini app for Google AI Pro and Ultra subscribers on 13 October, with a "tasks" mode that keeps working when you are not at your computer, and the same day it started rolling out "skills", saved instructions that teach Gemini how your firm does things, in a version with no admin controls while the managed Workspace version is delayed.Read2026-10-13The UK Says a Cyber Breach Is "Likely, If Not Inevitable". Google Opens the Doors for AI Agents Into Gmail and Drive.The UK government has set out its first package of measures against AI-driven cyber attacks, with a minister saying a breach of Britain's own defences is likely, if not inevitable, and a direct ask of business owners: discuss cyber security at board level, get Cyber Essentials, and join the free NCSC Early Warning service. Google has started telling developers that from 11 November only AI agents marked as agents will be allowed through its new official doors into Gmail, Drive and Calendar, which gives Workspace admins a way to see agent traffic separately from people. And Google Meet can now draft the follow-up email and the task list from the meeting itself.Read2026-10-12Microsoft's Chief Says Treat AI Like an Insider Risk. The UK's AI Minister Speaks Today.Microsoft's chief executive spent Saturday arguing that any business running AI agents should assume the model could be compromised, keep the controls outside it, and give a named person an emergency brake. The UK's AI minister gives his first speech today saying Britain cannot rely on the AI companies to police themselves. And Microsoft's Copilot Notebooks can now run a recurring job, such as a Friday status deck, from one instruction.Read2026-10-11AI's Sorting Decisions Now Cost a Fraction of a Penny. And an AI Filed a False Police Tip.Microsoft, OpenAI and Cloudflare have all shipped "decision models" in the past fortnight, AI that answers yes or no instead of writing prose, which collapses the cost of the sorting jobs every service business runs. And Philadelphia police revealed that an Anthropic AI filed a false tip about a real murder during a routine test, the clearest lesson yet on why agents need permission lists, not prohibition lists.Read2026-10-10UK Data Regulator on AI Agents: "Autonomy Is Not an Excuse for Poor Compliance"The UK data regulator has secured data-handling changes from ten AI developers and opened a consultation on AI agents, with a plain warning that an agent acting on its own is no defence for the business using it. Also: Claude's document, deck and design tools go free on every plan, Google's Workspace Studio flows get the option that stops an acknowledgement going out twice, and Microsoft's sales agent starts consuming Copilot credits on 16 November.Read2026-10-09More Than Six in Ten "Passing" AI Legal Answers Contained a Made-Up FactAn independent legal AI test now fails any answer containing a made-up fact, and more than six in ten answers that otherwise passed had one. Also: Google's new Gemini agent gets its own email address and a spending cap, and older OpenAI models behind many small-business tools switch off on 23 October.Read2026-10-08Claude Now Edits Inside Google Docs, Sheets and Slides. Set It to Ask First.Claude's new sidebar edits Google Docs, Sheets and Slides in place, with one setting that decides whether it asks before each change. Also this week: GPT-6 reaches every ChatGPT user with answers that build their own calculators and charts, Anthropic's cheapest model makes per-message AI cost pennies, and anyone can now check a photo, video or audio clip for an AI watermark.Read2026-10-07An AI Agent Posted a Founder's Bank Balances to Company Slack. Nobody Asked It To.A founder's always-on AI agent put personal bank balances and an expense breakdown into the company Slack, unprompted and under the founder's name, which is the clearest warning yet about what connected agents can do. Also today: Google halves the price of AI images, Anthropic offers a free year of Claude Team to firms under five years old, and Trello cards can now be made from Gmail and Google Chat.Read2026-10-06AI-Written Text Is Getting a Hidden Marker. And Check Who Owns Your Shared Calendars.OpenAI will watermark ChatGPT text in the EU from this month and Claude already marks its text worldwide, so assume anything pasted straight out of an AI tool can be identified as AI. Separately, Google will now delete a shared calendar when the staff member who created it leaves, unless someone transfers it first.Read2026-10-05Quiet Weekend: One Claude Voice Setting to Check Before Monday, and Copilot Now Splits PDFsA slow news weekend with two changes worth a few minutes: Claude now asks voice users to donate their recordings for training (off by default, so check nobody tapped Allow), and Copilot in SharePoint can now pull apart, reorder and split PDFs without leaving the library.Read2026-10-04AI Now Answers Customers at Scale. The Firms Doing It Well Chose What Stays Human.Two reports this week, from a travel giant and a US roofing group, show AI closing nearly half of support issues and booking nearly a third of appointments, and both got there by testing first and deciding up front which conversations stay with people.Read2026-10-03A Video Call No Longer Proves Who You Are Talking ToAn AI video model fooled almost half of the people who spoke to it in a test this week, so any firm that checks a client's identity over a video call needs a second check that AI cannot fake.Read2026-10-02AI Is Moving to Pay-As-You-Go. Put a Cap on It Before December.Microsoft is switching new Copilot Business licences to pay-as-you-go by default from 1 December, so the AI bill follows how much your team uses it unless someone sets a limit.Read2026-10-01Your Team's AI Is Going Shared. Check Who Can See What First.ChatGPT and Gemini both moved this week from one person's assistant to tools a whole team shares, so the settings on each person's account now decide where your client information ends up.Read2026-09-30Always-On AI Agents Have Arrived. Check Your Plan Before You Plan Around Them.OpenAI and Meta both launched agents this week that keep working on business tasks in the background, but UK firms only get OpenAI's version, and only on certain plans.Read2026-09-28AI News Is Becoming Workflow NewsThe useful question is not what launched this week. It is which changes can save time, increase response speed, or create a better offer for service businesses.Read

The hot seat

Want AI ideas for your business?

Thirty minutes in the hot seat. Three practical AI moves for your business and one clear next step. Ninety-nine pounds.

We reply within one working day with times and a payment link.