If I would’ve trained a markov chain on a set of cyberpunk novels, and other texts with similar context, I’m pretty sure I would’ve also gotten an “AI” that would be ready to kill and blackmail, lol.

yea that’s not how these models work, at all. they don’t act autonomously or just take over sentient-like control of everything that runs it, unless you tell it to. and probably wouldn’t be able to do that anyway without a lot more information it doesn’t have.