Or they could, you know, put some very basic logical safeguards in place to prevent their “super duper dangerous” AI from trying to hack real targets.

You know, just real incredibly basic things you do when you’re pentesting with scanning tools in a beginners lab type stuff.