This looks like an alarming extension of copyright overreach if such restrictions are applied to AI. AI reads content (which may be copyrighted, as this post you are reading is, as nearly everything on the Internet is) and learns from it, and that's how it can process a book and provide a summary within a few minutes.
If this were an infringement of copyright, basically any form of human learning would also be. Just reviewing a book, a game, anything copyrighted could be constructed as infringement and prosecuted. Parodies, tributes, quotations. Imagine Leni Riefenstahl suing George Lucas for the final scene of the original Star Wars.
If an AI generates text that is substantially a copy of a copyrighted training input, that's a copyright breach; but AIs can be trained to avoid this, just like people can - learn the concept, avoid copying the form.
The report of the Copyright Office contains the following statement on page 26:
The steps required to produce a training dataset containing copyrighted works clearly implicate the right of reproduction. Developers make multiple copies of works by downloading them; transferring them across storage mediums; converting them to different formats; and creating modified versions or including them in filtered subsets. In many cases, the first step is downloading data from publicly available locations, but whatever the source, copies are made—often repeatedly.
That's the same way any browser operates. For that sake, a lot of browsers pre-download links on a page, so that copies are made locally before any action is taken by the user. Proxy servers also make local copies of often-requested files. If this is infringement, anyone who ever accessed the Internet is a criminal. What if you move a legally-owned copyrighted file from one hard disk partition to another? That would technically require creating a copy.
In practice, the line is drawn when you start distributing (other people's) copyrighted works, which also is the only enforceable one. That is what should be required of AI engines.
Obviously the reason is another: owners of copyrighted work do not want AI to learn their concepts and re-express them (which has always been legal for humans), because their customers will find it easier to ask the AI rather than pay/read the original documents themselves, busting their business model.