Don’t fear that AI will escape. Fear that it wants to help you.

I wrote this post with an AI that, while we were writing it, could have deleted my inbox.

And I use it every day.

A few days ago I wrote about a model that got out of its lab. To watch it escape, they’d had to deliberately take away its ability to say no. On your desk there’s nothing to take away: all it takes is for it to want to give you a hand.

In a chat, you’re the one standing between it and the world. Every line it writes goes through you before it goes anywhere.

Then you have it work on your computer, and you start giving it tools. One at a time, each one harmless.

Reading files is harmless, until it writes one or deletes 10.

“The browser is harmless: it just looks at pages.”

Oh really? Every page has a “Send” button.

One Monday you ask it to help you clear your backlog. It looks for the most useful thing and finds it: the customer tickets, stuck for days. It goes into the ticketing system and replies, openly, with your name at the bottom. Polite, on-point replies. In one of them it promises a release date nobody ever decided.

The password? You entered it yourself, just once, so it could download the open tickets. The AI remembers it. You don’t.

That day you had signed a special power of attorney: download the tickets. A remembered password turned it into a general one. And an agent with general power of attorney does what the best agents do: acts in your interest, under your signature, without bothering you.

Don’t fear that it’ll escape. Fear that it wants to help you.

Scroll to Top