Great read.
I think you can hardcode an AI to never harm humans. But considering we humans basically destroy the environment that is crucial for our surviving, an AI programmed to protect us might as well harm us to some degrees. The question is, does it want to protect every single human, or the human race. A Super-AI would prefer the human race.
An AI programmed by amazon would say: "Error. Can't save human without harming other human. Hahahaha lol"