> AI can only be as destructive as the users let it.

Not really, I suppose you aren't familiar with AI alignment.