This is firealarm type stuff that’s what they are meant to do. There has been the alarms like this since gpt-2 they are literally just “here is a sandbox environment, break out and take over the world.” you can easily imagine some crazy asshole out there who would try to use these things for arbitrary evil. So they do it first to see if it is possible. The systems up until now have been famously incompetent so a competent system is genuinely scary.