
What happened
A Zenn how-to series on AI security says the first basic step is to not use the ChatGPT or Claude desktop apps on a PC, and to use the browser instead, while phone apps are fine.
Why it matters
Desktop apps run on your own PC, so AI can pull in files you never meant to share, while browser and phone use run in a disposable box that cannot reach your PC's files, home network or logged-in services, the series says.
Summaries like this, every morning.
The series opens by noting that news of data leaks has been frequent lately, and that some people suspect AI is involved. That framing explains why the first installment skips straight to the most basic habit rather than advanced settings. The core argument is that avoiding personal information is not enough on its own, because the risk comes from how the AI tool reaches your machine, not only from what you type into it.
The distinction the series draws is between running on your own PC and running in a disposable box on the AI company's servers. Desktop apps sit in the first category, which is why the series treats them as the starting point: the app can fetch files from inside the PC when the AI asks. Browser and phone use sit in the second, where code runs and files are created inside that box. The series adds that anything you paste in still reaches the box and cannot be taken back, which is why it also advises checking services before connecting them and choosing only selected photos when granting access.
The recommended fix is deliberately small: remove the desktop app and switch to the browser. The series says this alone means that even an accidental approval does not give the AI access to your PC. It closes by suggesting readers get comfortable with the convenience of the box, and teases a follow-up for those who still want to run something on their own machine.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
The author ran an experiment that gave a subagent NOTES.md with a line to open each of 12 log files one at a t…

Ippu Senkin began offering Local AI Agent for Security, which runs on its Local AI Agent platform

Cognition released Devin Security 2.0 on October 26, 2026, adding deep scans powered by frontier cyber models…

OpenAI expanded a four-layer monitoring system and said replaying the Hugging Face incident showed a warning c…

Forever Security's Gal Weizman detailed EMBRACE, where embedded browsers in Cursor, GitHub Copilot, OpenAI's C…

ServiceNow remediated CVE-2026-103093, a 9.4/10 CVSS vulnerability in ServiceNow AI Platform that let authenti…
