Given no information, would you assume companies that are buying "huge quantities" of "rare" books are getting first edition Jules Verne by the crate, or 1989 Highlights magazines recovered from dentist offices?

Both. I think they are just buying whatever they can get their hands on most of it is going to be junk, but one in a million is going to be a rare and valuable book which we didn’t even know existed.

After scanning and destroying 10 million books, AI companies will have destroyed 10 such books.

(I am obviously assuming a Poisson distribution here where I pulled the parameter p = 1/1000000 out of thin air; point is p is non-zero; and at this scale the undesirable event is bound to happen a bunch of times).