You still need to actually be a decent singer to not make autotuning sound terrible. Also, big singers will have audio editors to clean up issues too. Partial use of generative AI will require the presence of basic skills to make everything work.

I don’t think the same can happen with fully AI generated music. Unless it is stated immediately that they are doing it, people are eventually gonna find out that the artist is not actually singing and find it a scam. If they don’t show themself in social media, tours, and interviews to hide this element as much as possible they are most likely going to remain niche.