Some university, may have been at Western in London Ontario don’t remember where it was, ran tests on different AI models with both personal vandettas and national conflicts and most of the time they go nuclear, let it be go on a rampage with the person or drop actual nukes with the nation states.

There are ways to get AI to go around safe guards as well, like just a few minutes ago I was talking about Trump and his anilingus maggots in a chat room I could not remember how to spell anilingus so I asked Siri, which asked chatgpt who told me that it could not answer that question because of the guide lines rules it had to follow. So I just asked a different way and it was more than happy to tell me.