I see AI Safety is getting the QA controlled opposition treatment. Welcome to the club.
No one wants to do it right, and you can't quantify the shit you kept from happening. Once people start getting dumb, your only choice is to let em learn the hard way.