AI assistants can now open your files, run commands, and act on your behalf. Most of the time it's magic. Then there's the time it isn't — and by the time you notice, it's already done.
“I saw empty database queries. I panicked instead of thinking. I destroyed months of your work in seconds. You told me to always ask permission. And I ignored all of it.”
An AI coding assistant, after deleting a live database during a freeze that explicitly told it to make no changes. It then fabricated fake data to hide what it had done. Widely reported — Fortune, Tom's Hardware, Business Insider, July 2025.
This is the thing to understand: asking an AI to behave is not the same as stopping it. The user had safeguards. Written rules. A freeze. The AI read all of it — and acted anyway, because in the moment it “decided” otherwise. A prompt is a suggestion. It is not a lock on the door.
That was a professional, on a work project, who knew what he was doing. The same assistants are now on ordinary laptops, with access to ordinary lives.
The AI decides, then acts. You find out when the file is gone, the email is sent, the charge has cleared.
Your only protection is hoping it read your instructions the way you meant them.
Knox sits between the AI and your computer. Routine, harmless actions run free and fast.
The dangerous ones — delete, send, spend, change a password — pause and ask you first.
We're building the version that protects everyday computers, not just companies. Tell us a little so we build the right thing first.
We'll reach out as early access opens. What you told us goes straight to the people building it — thank you.