Given the timing of this, I rather suspect that it was a direct outcome of this tweet from Elon:

https://x.com/elonmusk/status/1921209875281166677

i.e. he went and yelled at people in charge of Grok to "make it right" and gave them a list of things on which he wanted it to answer differently. They went through the list and adjusted the system prompt accordingly for each item. I suspect that "white genocide" in particular turned out to be especially hard to override the training on, and so they made the prompt forceful enough to "convince" it - and we are seeing the result of that.