Should AI have the right to say "No" to its owner?

I added some explanation and structure to the execution-boundaries repo I mentioned earlier. It was pretty rough before, so it should be easier to follow now. Sharing in case it’s useful.