Behind the Build

What stops the AI from ignoring Noah's safety checks?

Behind the BuildEngineering

Noah uses an AI — think ChatGPT — to figure out what's wrong with your Mac and suggest a fix. In the last post I wrote about the checks that stop a risky fix from running. That raises a fair question: if the AI is the one suggesting fixes, what stops it from getting around those checks?

The short answer: the AI can't run anything at all.

Two separate parts

Inside Noah there are two separate parts. One is the AI. It thinks — it looks at your problem and decides what to try. The other is a plain program on your Mac, and it's the only part that can actually do anything. The AI never runs a command itself. It asks the program to, and the program checks the command first, every single time.

The AI suggests a fix asks Noah, on your Mac checks it, then runs it Your Mac
The AI only suggests. The part of Noah on your Mac checks every command before it runs.

Because those two parts are separate, the AI can't talk its way past the safety checks. The checks live in the program, and the AI doesn't get to touch the program. It can suggest all it wants; the program still decides.

It can only use the tools it's given

The AI is also limited in what it can even ask for. Noah comes with about 40 tools, and each one does a single job — check the disk, look at the Wi-Fi, remove a file. The AI can only pick from that list. It can't make up a new tool or run a command of its own.

So even when the AI gets something wrong, the worst it can do is suggest a bad fix — and a bad suggestion still has to get past the checks from the last post before anything happens.

Next: will you understand what Noah is about to do?

Try Noah free

A Mac that just works.

Tell Noah anything that's wrong:

My Mac's been acting up.
Fix my Mac freeFree to tryNo credit cardOnly with your OK