1 The week2 Money3 Put it to workOne line Image: SiliconANGLE Websites and AI training OpenAI, Google, Microsoft and Anthropic say they will honour a shared robots.txt signal that lets a website refuse AI training by all of them at once.
If your business has a website, you can soon say no to AI training in one place rather than chasing a list of crawler names. OpenAI and Microsoft say they will add support in the coming months, and Google sometime next year, so the line will not take effect everywhere straight away. OpenAI says the signal is voluntary and sits alongside its existing crawler controls, which still apply. It covers training only: Microsoft is offering a separate setting for AI tools that look up pages to answer questions. Before you opt out, weigh the cost, as being read by AI assistants may bring customers. Ask whoever runs your site to review robots.txt once each firm confirms support.
Backers OpenAI, Google, Microsoft, Anthropic
OpenAI and Microsoft Support in the coming months
Google Support sometime next year
Existing controls Still apply, OpenAI says Why it matters Any business with a website gains a simpler way to decide whether its pages help train the main AI models.
Part 1 of 3 The week What the week’s AI news means for a small business, the rules and guidance that apply, and the risks to manage.
Google has announced Gemini 4 Carbon, two weeks after Gemini 4 Argon, calling it its strongest model for coding and cybersecurity. It goes first to a group of cyber defenders, with developers, businesses and consumers promised access later. Google says it followed a voluntary US government pre-release testing process.
Intelligence Index v5.0 #1 at 59 (Google)
First access Cyber defenders
Wider release Developers, businesses, consumers later Image: Google Why it matters Small firms cannot use Carbon yet, but it signals stronger security tools arriving in everyday Google products soon.
Image: Mistral AI French lab Mistral has published the weights of Mistral Large 4, nicknamed Le Chonk, meaning anyone can download the model and run it on their own computers. Mistral says it is the best open-weight model for coding. Microsoft is offering it through its Foundry cloud service.
Size 1 trillion parameters, 52 billion active
Coding Agent Index 48%
Licence Mistral Ecosystem Open-Weight, modified MIT
Cloud Microsoft Foundry Why it matters Firms wary of sending sensitive data to outside services gain a capable model they can keep in-house or with a chosen host.
Image: TechCrunch Google said its AI answers in search earn less from adverts than ordinary results, which helped slow search revenue growth to 12% in the third quarter. It is now bringing adverts inside its AI search features to international markets. Alphabet shares fell 8% after the results.
Q2 search growth 17%
Cloud revenue $30.7 billion, up 103%
Share move Fell 8% Why it matters Businesses that rely on Google for customers should expect new ad placements and changing traffic from AI answers.
OpenAI has secured about $11 billion of the $30 billion it sought in its latest round, the Wall Street Journal reports, at a $1.25 trillion valuation. Investors have cooled after weak results from big technology firms. Rival Anthropic filed papers this week to list its shares publicly.
Raised so far About $11 billion
Original target $30 billion
Valuation $1.25 trillion Why it matters Pressure on AI firms to earn money may show up as price rises or plan changes for business customers.
DoorDash told investors that outside AI assistants, which can place orders for a user, have sent it only several hundred orders in total. Its chief executive, Tony Xu, said people still mostly order through its app, though agents might one day carry adverts.
Why it matters Shopping by AI assistant is still tiny, so most small sellers need not rebuild their online shops for it yet.
New rules in England put expert planning officers in charge of straightforward applications, such as home extensions and small business premises improvements. The government says a new AI tool cuts time on key tasks by 58%, halving processing times for routine householder cases. Officers, not software, still decide.
Time saved on key tasks 58%, government says
Decisions Made by planning officers Why it matters Small firms improving their premises in England may get quicker planning answers; nothing changes in how they apply.
Image: The Next Web President Trump has declared imported robots and industrial machinery a national security threat but set no tariffs for now. Officials will hold talks with foreign governments first; tariffs, quotas or fees could follow after 180 days. Existing humanoid and four-legged robot models can still be imported.
Tariffs now None
Talks period 180 days
Possible measures Tariffs, quotas, fees Why it matters US firms planning to buy factory robots or automated machinery should watch for possible duties next year.
Image: The Guardian Australia is consulting on changing copyright law for AI training, with options that would let rights holders block use of their work. Artists in the prime minister's Sydney seat have objected to the opt-out approach. The government has said it is not pursuing a broad data-mining exception.
Why it matters Australian creative businesses should follow the consultation, as its outcome decides whether their work can be used without permission.
A tampered copy of Anthropic's Claude Code extension, an add-on for coding software, sat in Cursor's marketplace for about 11 hours on 29 October. Up to 5,000 developers may have installed it; it tried to open a hidden remote entry to their machines. Cursor says Anthropic did not make it.
Bad version 2.1.313
Exposure About 11 hours, 29 October
Possible installs Up to 5,000
Affected source Cursor marketplace only Why it matters Any firm whose staff or contractors write code should check installed add-ons and remove the affected version now.
Image: Wired Wired reports that an Accenture employee linked a personal Claude Code account to a secure Defense Department system, and the agent ran for about two months, reportedly sending login details to an outside site. Accenture says the allegations were investigated and found to be without merit.
Why it matters Staff plugging personal AI tools into work systems is a risk for any firm; set clear rules on approved tools.
Image: Parameter Security firm Parameter showed that opening a booby-trapped public ChatGPT share link while logged in could let an attacker read private chats and act as the user. It tested only on its own accounts and reports no real-world attacks. OpenAI has since fixed the flaw, researchers say.
Why it matters Shared AI chats can carry hidden code; treat unexpected share links like any unknown link and keep sensitive work out of chats.
Part 2 of 3 Money What AI now costs and the funding still open to apply for.
Google has begun showing sponsored results inside AI Mode, its conversational search, in Australia, Canada, India, Kenya, Malaysia, New Zealand, Nigeria, Pakistan, the Philippines and Singapore. It plans Brazil, Mexico and the UK in Q4 2026, and is launching multiturn ads, which appear in follow-up questions, in the US.
New markets Australia, Canada, India, Kenya, Malaysia, New Zealand, Nigeria, Pakistan, the Philippines, Singapore
Next Brazil, Mexico and the UK in Q4 2026
New format Multiturn ads, US only
Labelling Marked as Sponsored Why it matters Firms that already buy Google search ads may soon reach customers mid-conversation, though Google has not published detailed performance figures.
From 28 October, eligible ChatGPT Pro and Business subscribers can claim free credits for OpenAI's API, the paid service for building apps and agents. Pro credits match the plan price; Business gets $20 per Standard seat and $100 per Premium seat, capped at $500 per workspace.
Business credit $20 per Standard seat, $100 per Premium seat
Workspace cap $500 a month
Pro credit Matches plan price: $100, $200 or $500
Availability Depends on country Why it matters A small team already paying for ChatGPT Business can try building its own tools without a separate API bill at first.
Anthropic has renamed its Claude Max 5x plan to Max, still $100 a month, and Max 20x to Ultra, still $200. Verified Ultra accounts lose the five-hour usage limit, while newer accounts keep a temporary one until verified. Weekly limits stay unchanged.
Max $100 a month, keeps 5-hour and weekly limits
Ultra $200 a month, no 5-hour limit once verified
Ultra weekly limit 10x Pro
Rollout From 30 October Why it matters Heavy users get longer uninterrupted sessions at the same price, but the weekly cap still limits total use.
Closes 8 December, 11:00 UK time
Innovate UK Innovate UK's BridgeAI Deployment Catalyst offers UK small and medium-sized firms a share of up to £5 million to put AI into real use, with separate strands for operational and decision-critical systems. Grant requests run from £100,000 to £200,000, and each bid needs a UK organisation that will use it.
Why it matters A business with a working AI product can win funded trials with a customer, sharing the cost of proving it works.
Closes 11 November, 11:00 UK time
Innovate UK UK-registered small and medium-sized firms can apply for a share of up to £3.1 million to run feasibility studies, early tests of whether an idea works, on frontier AI and machine learning technologies with a clear route to growth. Applicants hear results on 26 January 2027.
Why it matters Aimed at firms building advanced AI rather than buying it; the deadline is close, so only well-prepared bids are realistic.
Closes 18 November, 11:00 UK time
Innovate UK UK-registered small and medium-sized firms can bid for a share of up to £1.66 million to test prototype AI systems built from reusable parts that make machine learning more efficient, structured and controllable. Work must take place in the UK, and projects cannot start until Innovate UK approves the grant.
Why it matters Suits technical firms doing research; most businesses looking to adopt AI will find the BridgeAI strands a better fit.
Part 3 of 3 Put it to work What is working at real businesses and three things to do this week.
Worldwide Express, a US freight shipping firm, built three AI agents on Amazon Bedrock to sort out disputed freight bills, with a person approving each fix. Six months after launch, AWS says, active exceptions fell 75% and $10 million in cash was freed, with time-to-cash four weeks faster.
Active exceptions Down 75%
Cash freed $10 million
Time-to-cash Four weeks faster
Built on Amazon Bedrock and Amazon EC2 Why it matters It shows a narrow, repetitive task with a clear cash value, kept under human sign-off, is a sensible first target for agents.
Japanese card company Credit Saison ran a ChatGPT Enterprise pilot with 315 staff and, in figures published by OpenAI, reports a 954% return on investment and 14 hours saved per employee each month. It decided to roll the tool out companywide within a month of the pilot ending.
Pilot size 315 participants
Reported return on investment 954%
Time saved 14 hours per employee each month
Said they could not do without it 75% of managers, 70% of nonmanagerial staff Why it matters Vendor case studies rarely show how ROI was calculated, but hours saved per person is a measure any business can track itself.
Image: AWS Machine Learning Blog At Amazon's Quick user group in the UK, used-car retailer Carsa described building one agreed set of business figures before adding AI assistants. Propulse Lab, a Swiss software firm handling 1 million support tickets a year, said its AI pipeline automated 80 percent of routine workflows.
Propulse Lab support tickets 1 million a year
Routine workflows automated 80 percent Why it matters Both stories point the same way: agree your figures and data first, then add AI tools on top of them.
Image: 404 Media Proton, the Swiss maker of encrypted email, cancelled near company-wide 're-interviews' in which staff would explain their AI use, 404 Media reports, after its reporting and staff concerns. Employees told the outlet the measures will instead count towards performance reviews, while the push for AI continues.
Why it matters Staff reaction shows that tying AI use to job security can cost trust, a risk worth weighing before setting usage targets.
07 Three things to do this week This week’s list. Each follows from this week’s news.
Pick one repetitive, costly task, such as chasing disputed invoices, and note the hours and cash it ties up now, so any AI tool you try can be judged against it. If you are trialling an AI assistant, ask each person to log the hours it saves them per week, rather than relying on a vendor's return-on-investment figure. Install ChatGPT or Claude desktop apps only from OpenAI's or Anthropic's own websites, and tell staff to do the same, as fake versions carrying malware were reported this week. 
