The same thing, legally, according to the judge in that lawsuit would have been a purchased (potentially used) copy of the book scanned - i.e. what Anthropic also did after pirating works. It'd be surprising if that would cost even $30USD/work, two orders of magnitude less.
$10AUD million doesn't seem sufficient for a competitive set (and as you say they aren't saying it is), but if you told me $50AUD million was enough to build a legal (according to Judge Alsup's interpretation of US law) repository of training data I would not be surprised.
The millions and billions you hear about in copyright "settlements" are just the amount it takes to bribe a local court, so $10MM is reasonable for Australia.
Not to defend the absurd statement about a bribe, but with regards to the $1.5 billion dollar settlement this isn't quite true. Because it's a class action the judge will also have to approve the settlement - finding that it was fair and agreed to without collusion. This is done because the incentive structures set up a bit of a conflict of interest between class action lawyers and members of the class... Of course none of the settlement goes to the judge or court, there's no bribing going on. But judges do reject class action settlements sometimes.
> June 24 (Reuters) - A federal judge in San Francisco ruled late on Monday that Anthropic's use of books without permission to train its artificial intelligence system was legal under U.S. copyright law. Siding with tech companies on a pivotal question for the AI industry, U.S. District Judge William Alsup said Anthropic made "fair use", opens new tab of books by writers Andrea Bartz, Charles Graeber and Kirk Wallace Johnson to train its Claude large language model.
https://www.reuters.com/legal/litigation/anthropic-wins-key-...