It’s worse. They want to train their models on data their competitors cannot access. So they buy rare books, scan them in, and destroy the source material.