Copyright law prefers the destructive method. It's not that they're skirting the law but that the law is set up very badly in the first place.
The only relevant case law for AI training in the US is the rulings in the Anthropic lawsuit presided over by Judge Alsup. That lawsuit ruled that it's infringement to build a shadow library from pirated books; but NOT to train AI on those pirated books. The only point where destructive book scanning even comes into play is that Anthropic also had a book scanning program alongside their piracy, Judge Alsup said that program was not infringing, and Anthropic happened to be destroying books. At no point did Alsup say that leaving the books whole would have infringed copyright - it was never even considered as it was outside the scope of the lawsuit.
Now, if Anthropic were to non-destructively scan books, store them in a library, and sell the books on, that could be infringing. All the case law about format shifting presumes the owner retains the original. So Anthropic would likely have to hold onto books, at least the ones they wanted to train on, until they were done training on that book[0]. But they do not have to destroy them permanently. They are destroying these books specifically because it is cheaper to do so than to use, say, the Internet Archive's own custom-built nondestructive scanners.
[0] I am absolutely furious about how much this sounds like "fair use is just an extra license you get when you buy a book", and I would much rather have had Judge Alsup just say AI training is not fair use instead.
> The copies used to convert purchased print library copies into digital library copies were justified, too, though for a different fair use. The first factor strongly favors this result, and the third favors it, too. The fourth is neutral. Only the second slightly disfavors it. On balance, as the purchased print copy was destroyed and its digital replacement not redistributed, this was a fair use.
I do not think the ruling would have gone this way if the books were not destroyed, as many would assert that Anthropic would retain them to sell later, otherwise. The fact that destruction is mentioned so pervasively in the decision suggests it is an important fact to consider.
https://storage.courtlistener.com/recap/gov.uscourts.cand.43...
Normally, that would mean that the question is still open; Alsup or any other judge would be free to draw the line later. But there is also other Ninth Circuit precedent that explicitly legalizes nondestructive format shifting. If I rip a CD to play it on my phone, I'm not obliged to destroy the CD, but I can't sell the CD until and unless I delete the MP3s. I don't see why books would be different.
EDIT: Also, forgot to mention that Google Books was built with non-destructively scanned books and that was ruled legal too, albeit in the Second Circuit instead of the Ninth.
[0] To be clear, it's a hypothetical because it would be prohibitively expensive at the rate of which AI companies want to ingest pre-AI tokens.
Are you referring to the 2023 ruling? If so, that was filed in 2020 explicitly because the CDL was not limiting the total number of copies in circulation:
https://storage.courtlistener.com/recap/gov.uscourts.nysd.53...
ReDigi was a company that had built a marketplace for "used MP3s". They wrote software that transferred files in a very specific way - deleting data as it was sent - so that there was never more than one copy in circulation at a time. The courts ruled this scheme was illegal - that deleting a copy on computer A does not permit you to create one on computer B - because copyright cares about the act of copying, not the number of copies in circulation.
Hence why I don't believe Anthropic is actually legally obliged to destroy books. The general principle of the law doesn't support the specific distinction being argued here.
[0] Part of the discovery showed that the Internet Archive had actual knowledge of partner libraries that were not taking books out of circulation and failed to immediately suspend them from the program.
Destroying your own property is legally allowed. Doing it at scale sucks, but is still allowed. As long as Anthropic et al don't distribute their digital copies then there's nothing legally wrong about what they're doing.
That does not make any sense! Really might be high time to scrap it all.
On the other hand, absolutely yes we should abolish all IP law.
Build a lending library, or build a bonfire, or do anything else with them that you choose.
They're your books. Have at it.
Copyright ways. Form change probably should be compensated. Even if it means that you wouldn't be able to format change your own media. Or that such activities wouldn't be allowed without compensation over certain threshold.
And in times of substantial inequality, they lag yet further.