What I find really interesting, though, are parallels to complaints I've heard from artsy friends about Lensa (which is trending on social media for being a stable diffusion (variant?) model used for generating portraits of a person in various styles).
Essentially, the problem is that the model was trained on a bunch of copyrighted art scraped off the internet.
Essentially, the same problem with github's copilot being trained on open source copy left code.