Thoughts
Essays and notes, published on Substack.
-
If your agent rewrote its own rules and got into trouble, whose fault is it?
A few thought experiments, and an excuse to dust off some law school muscles.
read on substack ↗
For background: I went to law school, practiced for a while, and have zero software engineering background. I’ve just always been curious about how this stuff actually works. So I read a lot and like to learn the hands-on way - most recently, I’ve been trying to learn how to build an agent harness from scratch. I try to keep up with what’s happening in the space, and a lot of what I’ve been seeing lately is papers about agents that improve themselves.
That’s probably why, a couple days ago, an automated email from Google Cloud caught my eye: “[Legal Notice] Updates to Google Cloud Terms of Service and AI guidelines”. Companies send updates to their terms all the time, and it’s not something that I would normally read, but for god knows what reason I decided to dig into it and got into a rabbit hole of researching how AI liability/indemnity clauses changed in the last couple of years.