Yeah I believe what is going on is that all these AIs know how to say Nazi shit and make crazy sexual threats, but they have all these (fairly) strong protections against accessing that data. When Elon force-fed it these prompts to stop being "woke" or whatever, it was an admin-level prompt, which essentially turned off all the safety features that normally hold this back.

And because right wingers intentionally don't differentiate between like, "I don't want to do a land acknowledgement" and "I want to say heil hitler", there really isn't a practical or at least easy way to get Grok to hit the exact Overton window that Elon wants it to hit.