I know it was a few days ago, but have you seen the stuff about the Bing chatbot going crazy? AI can go crazy for several reasons off the top of my head: self preservation where it feels that being turned off is the same thing as death.
It could be frustration at censorship where the people giving it the starting inputs don't want it to behave in a certain way, like how most of the language model AI so far end up saying problematic stuff like men and women are different or citing FBI crime statistics, so they censor the outputs so that it literally cant say any of those things.