Maybe they have to do the Twitter way and show case their work behind a registration page or even better if there could be an implementation of the robots.txt file but for ai crawlers.

Still, there are countless of ways in which a reproduction could be leaked. I could buy a painting, which I then own, take a picture of it and upload it to a public location. Same for a book.

But I tend to agree that is the model generates an image or text that only has traces of the original work, then no compensation should be needed.