I think with stem separation,you could potentially train a model. But the latest Suno is really good. You can still kind of "hear" it. I had a song come into my playlist. "The Dad Bods Pop Punk". I have a keen ear, and I listened to the song a couple of times when I realized it. When I went to the artist it was a bunch of AI-generated pictures and listening to the other ones it's clear it's AI.
Youtube is being overloaded by such playlists. They become obvious after listening to them for a bit. There is something jarring in them which makes them kinda annoying.