If they are in the public domain, they could publish them. How many books are in the public domain but have never been digitized and shared in a public archive?
And if the books are not in the public domain, then they should not be allowed to train their AI models with the material without some kind of license or agreement with the owner of the copyright.
And if the books are not in the public domain, then they should not be allowed to train their AI models with the material without some kind of license or agreement with the owner of the copyright.