The Group Chat · week of 2026-10-26

Before you say yes: what your AI tools can read

The Group Chat

Welcome to the first Group Chat. This week’s big idea is a small habit: every time an AI tool asks to see more of your screen or your messages, treat it like a permission slip, because the person signing it is usually you and your clients.

In this issue:

The Byte: Your AI helpers want to see more of your screen
The Byte: Using Claude for client work? Its new rules start Nov. 12
Club news: a spotlight, 2 rooms to be in
Study guide: Agentic AI, Browser extension, Cloud-based, plus a quiz

The Byte

Your AI helpers want to see more of your screen

Two updates this month point in the same direction. The Verge reports that ChatGPT can now look at the Chrome tabs you have open when you tag Chrome inside its web or desktop app, as long as OpenAI's Chrome extension is installed. In OpenAI's demo, ChatGPT compares three sofa listings sitting in separate tabs. Google, meanwhile, says an opt-in Enhanced Protection mode is coming to Android's Messages app over the next few weeks. When someone outside your contacts texts you, Gemini models running in the cloud, meaning on Google's servers rather than on your phone, can read that conversation to judge whether it's a scam. Both features solve real annoyances. Comparing supplier quotes or competitor pages side by side is tedious, and scam texts cost time for anyone who takes work over the phone. The catch is the same in both cases: the help works because the AI sees more. A tab with a client invoice or your inbox open may be visible to ChatGPT once you tag Chrome. A first text from a new customer asking about a quote is exactly the kind of message Google's AI would read. Expect this tradeoff to keep showing up. OpenAI also mentions bringing "computer use," which means AI that operates the browser on your behalf, to Chrome users, though the details aren't public in the reporting so far. The habit worth building now is treating every new AI permission as a decision about whose information you're sharing, and for people who work with clients, that usually includes theirs.

Byte's take: This matters most if you run your business through your browser and your phone, since client details live in both. If you only use ChatGPT in its own window for writing help, nothing changes for you yet.

Your move: Type chrome://extensions into Chrome's address bar and read what the ChatGPT extension is allowed to access, if you have it installed. Before you tag Chrome in ChatGPT for the first time, close any tab showing invoices, client files or email.

Using Claude for client work? Its new rules start Nov. 12

Anthropic published a revised Usage Policy for Claude on October 8, and it takes effect November 12. A usage policy is the company's rulebook for what you're allowed to do with its product. Anthropic says most of the changes clarify rules that already existed, but two sections deserve a read if Claude touches anything you get paid for. The first is the High-risk Use Case Requirements section. If Claude shapes decisions about someone's health, finances, legal rights or livelihood, Anthropic already required a qualified person who can change its recommendations, and the affected person has to be told AI was used. The rewrite now spells out which uses count. If you help clients with budgets or contracts, that list is how you find out whether the disclosure rule applies to your work. The second is a new deceptive campaigns section, which pulls existing rules against fake accounts and messaging with hidden sponsors into one place. Anyone running social media or marketing for other people should know it before drafting a batch of posts with AI. Anthropic also added a rule against sustained cruelty toward the model, and says it excludes ordinary frustration and dark creative themes, so snapping at a bad draft doesn't count. The larger pattern is worth noticing. As AI moves into real client work, the tool maker's rules start to shape what you can promise the people paying you. Checking them before a deadline is far easier than untangling a problem after one.

Byte's take: This matters if you use Claude for client deliverables, especially anything involving money, legal questions or marketing campaigns. If you only use it for personal brainstorming or learning, a quick skim is enough.

Your move: Before November 12, read the High-risk Use Case Requirements and deceptive campaigns sections of Anthropic's updated policy and write down any client project that falls under them. For any that do, plan how you'll tell the affected person that AI is part of the work.

Club news

Spotlight

Next week’s Spotlight could be you: reply with something you built, fixed or figured out with tech this month, and we’ll share it, no name unless you say so!

Rooms this week

Microsoft Excel basics class, Thursday, Oct. 29, 10:30 to 11:30 a.m., Detroit Public Library Parkman Branch (1766 Oakman Blvd.), free. Register with the class QR code in DPL's Fall 2026 tech catalog. Details
Halloween Cards with Canva · Wed, Oct 28, 2:30 PM–3:30 PM · Main Library: TLC Computer Lab, 5201 Woodward Avenue, Detroit, MI 48202, Detroit · Free Details

Study guide

Agentic AI: Software that can take actions for you, such as clicking through a website or filling in a form, besides answering questions. OpenAI says it plans to bring "computer use," meaning AI that operates your browser for you, to Chrome users, which is this kind of AI.
Browser extension: A small add-on you install in a web browser like Chrome that gives it extra abilities. ChatGPT's Chrome extension can now read your open tabs when you tag Chrome, so it's worth checking its permissions in Chrome's extensions settings first.
Cloud-based: Done on a company's computers over the internet, so your information leaves your own phone or laptop to be handled there. Google's new opt-in scam screening in Android Messages sends texts from unknown numbers, including possible new clients, to Google's servers for its AI to read.

Quick quiz

Which of these is an example of agentic AI?
A) A chatbot that explains what an invoice is
B) An assistant that opens your browser and fills in an order form for you
C) A spell checker that underlines typos
You just installed a ChatGPT add-on in Chrome. What is a smart first step?
A) Review the extension's permissions in Chrome's extensions settings
B) Open all your client invoices so it can learn your business
C) Turn off your computer's password
If a scam filter is cloud-based, what happens to a text from an unknown number?
A) It stays only on your phone and is never shared
B) It gets deleted before you can read it
C) It is sent to the company's servers to be checked

Answers: 1. B, Agentic AI takes actions on your behalf, like operating a browser, which goes beyond answering a question. 2. A, A browser extension can see what its permissions allow, so checking them tells you what it can access before you rely on it. 3. C, Cloud-based means the work happens on the company's computers, so the message leaves your phone to be read there.

Got a word you keep hearing? Send it in: drop it in the chat and pick "A word".

What’s one thing you’d never let an AI see, even if it promised to save you an hour?

How was this one? Worth the read · Some of it · Not for me