We all learned the old security rules. Close your browser windows, close your apps, assume your phone knows where you are. Most people who use a computer for work have internalized some version of that list, and it has served them reasonably well for two decades.
The AI era has a new rule that most people have never heard of. It has nothing to do with browser tabs or location settings, and everything to do with a category of file that quietly accumulates on your machine every time you open an AI tool.
What The Tools Keep
Every AI tool you use, from ChatGPT to Claude Code to Cursor, keeps a transcript of what you told it and what it touched. That is not a hidden feature or a dark pattern. It is how these products were built, and in a lot of cases the logging genuinely helps with debugging and recovering a session you lost.
The part that surprises people is the format. On most of these platforms that transcript is a plain text file sitting in a hidden folder on your computer. No encryption, no password, no special viewer required. Anything that can read a text file can read it.
Everything In The Transcript
Think about what actually ends up in there. Your prompts, which are often more candid than anything you would put in an email. The files the agent opened, which maps out your project structure and your client work. The passwords a command printed to the screen, because a terminal-based agent logs what the terminal showed it.
That last one is the one that stops people. Web Experts has watched enough agent sessions to know that credentials show up in terminal output more often than anyone would like to admit. Once they show up, they are in the transcript, and the transcript is a plain text file.
Why Attackers Care
Attackers are already going after these files, and the reason is simple economics. Breaking encryption is hard and slow. Reading a text file that a developer's own tooling wrote for you is neither of those things.
For a small business, this is the kind of exposure that does not show up on any compliance checklist yet. You can have good password hygiene, two-factor everywhere, and a locked-down network, and still be leaving a running narrative of your work sitting in a directory nobody thought to look at.
What To Actually Do
The first step is knowing where the files live. Each platform stores them in a different place, and Web Experts wrote up where each one puts them, why attackers are already going after them, and what to do about it. Once you know the paths, the rest is ordinary hygiene: review them, clear them on a schedule, and keep them out of anything that syncs to a shared drive or a backup you do not control.
The broader habit matters more than any single setting. The old rules made sense because people repeated them until they became reflex, and this one deserves the same treatment. If you are running AI agents on a work machine, assume they are writing everything down, because on most platforms they are.