Well, it did threaten violence towards a user.
If I was launching AI, having it be just a bit unhinged is obviously the best way to get it to go viral.
Sure, not your run of the mill stupidity, but even ML scientists do not understand why would a model make a particular choice.