Security Vignettes

No. 39|AI & agents

分享
39 / 43

The AI that moved into our computers

AI tools moved into our computers, and IT can’t tell who installed what.

Out of the drawers of an old apothecary cabinet crawl several little clockwork gadgets; the little robot inspects them by lantern light.

(Illustrative scenario, not a real incident)

A year or two ago,using AI meant chatting in a browser.

Not anymore.AI coding tools run commands on the computer;all kinds of small connector services run in the background;AI agents can read and write files,call services, and operate other programs.

At this twenty-five-person software company,each engineer had installed several tools,plus plenty of extensions found online.Some read the entire project directory;some had keys written right into their config files.

One day, word came:a very popular extensionhad been found to have a malicious version.

IT wanted to know if anyone at the company had installed it,and found there was no way to check.

What you see is higher efficiency.What you don’t seeis that those tools’ permissionsare the user’s own permissions.

The methods for inventorying cloud servicescan’t catch what’s installed on a computer.

I used to think AI lived in the cloud.Later I learnedit had long since settled into our computers,using the same keys we do.

Before you invite a guest in,at the very least, know who they are.

What you can do today
  • Ask engineers to list the AI tools and extensions they’ve installed.
  • Install only from trusted sources, and pin versions.
  • Move keys out of config files and fetch them from a secure store.
How did this make you feel?
分享