There are uses of AI that are proving to be more than black and white. While voice actors, have protested their performances being fed into AI against their will, we are now seeing an example of this being done, with permission, in a very unique case.
I like the idea a lot. I can think of two problems that need to be solved for this:
Multiplayer doesn't make a difference, at least not noticeably, as all those shenanigans can already be done by text. You'll get reported and named "player345133" whether you misspell e.g. a forbidden slur or use a misspelling to get the slur's pronunciation.
But it's harder to detect spoken slang when people make it talk in mixed accents and more, you would have to run a talk to text engine too many different times with different filters and parameters against many different languages' lists of slang words with multiple recognizable pronunciations.
And then somebody names themselves ChatGPT and the French will laugh because that sounds like "cat farted" in their language