They could use them as training data, without providing access the actual books.
This doesn’t incentivize them to be good stewards of this data and making anything in the public domain available. It incentivizes less access to the source material, having to blindly trust their tools, and is effectively automating plagiarism.