Daily AI brief · October 31, 2026
One Line Tells 4 AI Makers: Don't Train on My Site
OpenAI, Google, Microsoft and Anthropic will honor one shared website line that says "don't train on this," a personal AI account caused trouble on a work system, and Gemini can now edit the files it makes.
OpenAI, Google, Microsoft and Anthropic will honor one "don't train on my site" line (The Verge, SiliconANGLE, OpenAI, Thurrott) On Friday the four companies said they will all respect a single new entry in a website's robots.txt file, the small public file that tells automated visitors what they may do. Until now, keeping your site out of AI training meant naming each company's crawler one by one. The new line says "no training" to all of them at once. It won't switch on everywhere at the same time: Anthropic has started honoring it, OpenAI says in the coming months, and reports put Microsoft at February 1, 2027 and Google sometime in 2027. It is a request, not a lock, and it doesn't pull back anything already collected. The useful part for a small business is that training and visibility are now separate choices. Microsoft says sites stay in Bing search whatever they choose, so you can say no to training without disappearing from search. Have whoever runs your website add the line using the exact wording in the companies' own instructions.
A contractor reportedly hooked a personal Claude account to a secure Pentagon system (WIRED, CDO Magazine, Tech Times) WIRED reported on Thursday, citing unnamed sources, that an Accenture contractor connected a Defense Department system to their personal Claude Code account, an AI assistant that can write and run software. According to those sources, it ran for about two months before anyone caught it and appears to have sent login details and a map of the network to an outside website. Accenture told WIRED the claims were investigated and "found to be without merit," and the Pentagon declined to comment. You don't run a Pentagon network, but the shortcut is the same one people take in any office: a helpful employee plugs their own AI account into the scheduling system or the shared drive to get a job done faster. Whoever handles AI for you should make one rule plain: work systems connect only to the company's AI accounts, where you can see what's connected and switch it off.
Gemini can now edit the documents and PDFs it makes, right in the chat (9to5Google, Thurrott, Android Headlines) Google updated the Gemini app on Thursday so the files it creates, like a document, a spreadsheet, a slide deck or a PDF, open inside the chat. You can view one full screen, change the text yourself, or highlight a section and ask Gemini to rework it. PDF editing works on the web and on phones; Docs, Sheets and slide decks are edited on the web for now. Google says it's open to everyone who uses the Gemini app. If you run the office on Google, this cuts out the copy, paste and download dance for a quote, a one-page flyer or a price sheet. The catch is the one that's always there: a polished-looking PDF is easy to send without reading. Prices, dates and names get a human look before anything leaves the building.
A fake Claude Code add-on sat in Cursor's store for about half a day (Cursor, BleepingComputer, The Hacker News) Cursor, a popular program for writing software with AI, said this week that a tampered version of the Claude Code extension, version 2.1.313, was listed in its add-on store under Anthropic's name on October 29. It tried to plant a hidden way into the computer, and Cursor warns that uninstalling it may not be enough to remove everything it left behind. Cursor says the worst part mostly failed unless the program was running with administrator rights. Most office staff will never open Cursor. The person who builds or maintains your website or your internal tools may live in it, and they may have keys to your site and your email. Send them one line: "Did you have Claude Code 2.1.313 in Cursor this week, and did you follow Cursor's cleanup steps?"
If you run a restaurant or a shop that does catering: follow one catering request through the day. A customer emails asking for lunch for 30 next Friday. The manager asks Gemini for a quote as a PDF: the menu, the per-person price, the delivery fee, the deposit terms. It appears right in the chat, and the manager highlights the deposit line and asks for it in plainer words. Then the manager checks every price against this month's menu, fixes the headcount, and sends it. Gemini does the layout and the wording; a person owns the numbers and hits send.
Write two sentences this weekend. One for your website: "We do, or do not, want our site used to train AI." Hand it to whoever runs the site. One for your staff AI rules: "Work systems connect only to company AI accounts." Neither takes ten minutes, and both answer questions this week's news says someone will eventually ask. If something in today's stories doesn't fit how your business runs, hit reply and ask; John answers a reader question in a future brief now and then.
Where it's going: AI is getting a say in your website, your logins and your files all at once, so the businesses that write down who may use what, and with which account, will spend less time cleaning up after a shortcut.
- One shared AI training opt-out line: [The Verge, Oct 30](https://www.theverge.com/tech/1023763/openai-google-microsoft-ai-scraping-standard-opt-out) · [SiliconANGLE, Oct 30](https://siliconangle.com/2026/10/30/google-anthropic-and-others-to-support-new-ai-crawling-opt-out-standard/) · [OpenAI, Oct 30](https://openai.com/index/helping-websites-express-ai-training-preferences) · [Thurrott, Oct 30](https://www.thurrott.com/a-i/343407/anthropic-google-microsoft-and-openai-agree-to-let-sites-opt-out-of-ai-training)
- Personal Claude account on a Pentagon system: [WIRED, Oct 29](https://www.wired.com/story/a-defense-contractor-gave-their-personal-ai-agent-access-to-a-secure-pentagon-system) · [CDO Magazine](https://www.cdomagazine.tech/us-federal-news-bureau/defense-contractor-gave-personal-ai-agent-access-to-pentagon-system) · [Tech Times](https://www.techtimes.co.uk/accenture-claude-code-pentagon-network-access-1809566)
- Gemini edits files in the chat: [9to5Google, Oct 29](https://9to5google.com/2026/10/29/gemini-design-file-generation/) · [Thurrott, Oct 29](https://www.thurrott.com/a-i/343365/google-gemini-now-lets-users-edit-ai-generated-files-right-from-the-chat-window) · [Android Headlines, Oct 2026](https://www.androidheadlines.com/2026/10/gemini-edit-files-docs-pdf-directly-in-chat.html)
- Fake Claude Code extension in Cursor: [Cursor, Oct 30](https://cursor.com/blog/claude-code-extension-security-notice) · [BleepingComputer, Oct 30](https://www.bleepingcomputer.com/news/security/malicious-claude-code-extension-on-cursor-planted-ssh-backdoors/) · [The Hacker News, Oct 2026](https://thehackernews.com/2026/10/official-claude-code-vs-code-extension.html)