I suppose the whole regime. I'm not an AI safetyist, mostly because I don't think we're anywhere close to AI. But if you were sitting on the precipice of atomic power, as AI safetyists believe they are, wouldn't caution be prudent?
I’m not an expert, just my gut talking. If they had god in a box, US state would be much more hands on. Now it looks more like an attempt at regulatory capture to stifle competition. ”Think of the safety”! ”Lock this away”! If they actually had skynet US gov has very effective and very discreet methods to handle such clear and present danger (barring intelligence failure ofc, but those happen mostly because something falls under your radar).
For example: Two guys come in, say "Give us the godbox or your company seizes to exist. Here is a list of companies that seized to exist because the did not do as told".
After? They get the godbox. I have no idea what happens to it after that. Modelweights are stored in secure govt servers, installed backdoors are used to cleansweep the corporate systems of any lingering model weights. Etc.