This post was updated October 2.

You have likely heard reporting about the Anthropic settlement to compensate authors for the illegal use of books to train their Large Language Models for AI, like the Claude series. Note this is not payment for an ongoing license, but compensation for past copyright infringement. There are many cases with the US courts, against multiple AI developers who have stolen intellectual property for this purpose.

“Perhaps the most important takeaway from all this is what this settlement is not: this settlement doesn’t establish new precedents for AI training or create ongoing licensing obligations for Anthropic. It simply resolves liability for past conduct involving the LibGen and PiLiMi datasets, leaving the broader legal landscape around AI training largely unchanged. Judge Alsup’s earlier ruling that LLM training constitutes fair use remains intact and continues to provide important guidance for the industry.”

Alsup’s ruling sets a precedent that, provided you buy your first copy, then scanning and ingesting it for AI training is fair use – no-license-required. In other words – don’t pirate – but after that, please feel free to take works for your machine. Therefore the settlement doesn’t “weaken” the fair use defence – at least not for AI training.  It does weaken the fair use defence for piracy.  But as to AI training, the decision is the opposite and a major blow to rightholders:  Alsup decided the fair use defence for AI training squarely on the side of big tech. Rather than appealing this, by settling the plaintiffs allows Alsup’s fair-use decision to go unchallenged.  This is why Anthropic is settling.

This specific settlement has been rejected by the US courts as too narrow in scope. It would only pay authors and publishers who had books lodged with the US Copyright Office. A book is automatically copyrighted when you write/publish, but in the US there is a separate process (that costs money) to register copyright officially on titles, and under US law, that registration is required to make a copyright claim in the US. This meant that only around 500k of the $7.5 million books ingested would be compensated in this settlement, and that is being challenged by US writers and publishers who have not registered with that entity, and for books from other territories.

The settlement says that each author(s) work will share the approximately US$3,000 award with their publisher. This fact has been lost in some of the coverage of the suit. Based on the September 8 hearing, the judge has requested supplemental information from class counsel by September 22 that should provide clarity on the allocation between authors and publishers. The allocation should be based on actual contracts or at least on contract norms, because many contracts are silent on the allocation of infringement awards, and many historic contracts do not have AI licensing splits in the subsidiary rights schedules.

Here is where you register books with the US copyright office and Info on How and a
You Tube clip Also see the Created by Humans licensing website

The halt on the Anthropic settlement while it is being appealed still only currently only covers books registered with the US Copyright office by the date of that settlement. Registration might still be advisable for future settlements. NZSA is working with our sister orgs, the UK Society of Authors and the Australian Society of Authors, consulting with the US Authors Guild,  whose legal teams are working to establish if authors/publishers in other copyright territories could be compensated. Publishing Associations and Collective Management organisations (such as CLNZ) are also very much engaged.  A key meeting is taking place September 25 and we will update you.

This is however, a signal to businesses, government entities, companies et al, to avoid illegally produced LLM’s with cases before the courts, as they are not compliant under international copyright treaties NZ is signatory to, such as the Berne Convention, and therefore present a risk to business.
NZSA newspost link HERE

US Authors Guild post
Anthropic tells US judge it will pay $1.5 billion to settle author class action
AAP Endorses the US$1.5 Billion AI Settlement Proposed ‘Bartz v. Anthropic’ Case

AI news this Fortnight:
Government AI strategy only the start
Our AI strategy needs to include everyone, not just business
Bloomsbury AI Licensing deal
‘Horrifying and incredible’: one writer’s argument for making use of AI

UK Society of Authors message:

The Anthropic Settlement is a class action consisting of authors and publishers who claim that their copyright was infringed by Anthropic PBC, which used books from pirate websites Library Genesis (“LibGen”) and Pirate Library Mirror (“PiLiMi”) without permission to train their Large Language Models.

The settlement terms proposed by the Plaintiffs’ attorneys in the action being brought by U.S. authors against AI company, Anthropic PBC, were considered by the Judge, Hon. William Alsup, on 8 September.

The settlement was not approved, with the Judge seeking clarification from the parties on a number of points, including the works which will benefit from any settlement.  The Judge has asked for further information also to determine the split between publishers and authors.  The matter is due to return before the court on 22 September when the parties will be making further submissions in the hope of persuading the Judge that the settlement proposed is “fair, reasonable and adequate”. Book industry legal teams are then meeting September 25.

We will advise members of the implications of the settlement for authors if it is approved. One of the key issues will be whether the works of UK/Australian/NZ authors fall within the definition of works which will benefit from the settlement. As currently drafted, the definition provides that the settlement would apply to:-

“All beneficial or legal copyright owners of the exclusive right to reproduce copies of any book in the versions of LibGen or PiLiMi downloaded by Anthropic as contained on the Works List. “Book” refers to any work possessing an ISBN or ASIN which was registered with the United States Copyright Office within five years of the work’s first publication and which was registered with the United States Copyright Office before being downloaded by Anthropic, or within three months of publication. Excluded are the directors, officers and employees of Anthropic, personnel of federal agencies, and district court personnel. For avoidance of doubt, only works included on the Works List are in the Class.”

The Judge has expressed concerns about the parties’ proposed approach to the Works List and has invited the parties to submit a final Works List to the Court on Monday for consideration.

Given the definition of “Book”, the class of books which will be covered by the settlement is effectively closed and there is no benefit in authors now registering works with the U.S. Copyright Office for the purpose of sharing in this settlement; if a book qualifies it should appear on the Works List which is currently being finalised.  However, if this settlement serves as a model for the settlement of other legal actions which are being brought against AI companies, registration of works with the U.S. Copyright Office may be beneficial if those books have been downloaded or torrented from an unlawful dataset.

The Works List is being compiled by reference to the raw text of the files that Anthropic downloaded, the accompanying metadata, data from the United States Copyright Office and the International Standard Book Number (ISBN) and book metadata from Bowker, the key industry source of such information.  There is, therefore, no need for our members to send any information to the Plaintiffs’ attorneys about their books.

The settlement terms anticipates that once the Works List is finalised, a searchable database will be made available so that authors can check whether any of their books have qualified.  If a book is on the Works List, then authors will be invited to contact the Settlement Administrator to ensure that they have their current contact details so that (i) they can be invited to comment on the proposed terms or opt out; and (ii) they can participate in the settlement.

We will be providing regular updates to members in relation to future developments as they occur.

Read the Author’s Guild ‘What Authors Need To Know’ for more details.

Summary:

The class definition currently says only US editions that were registered with the US Copyright Office at the time of the infringement will be included.

1. It is not compulsory to register copyright with the Copyright Office. ISBN’s are not tied to this registration – this is a separate process and attracts a fee.
2. Not all publishers have been registering books with the US Copyright Office, even though in some agreements/contracts with authors it was stipulated that they would have.
3. Publishing contracts do not often have splits in rights for legal settlements, and many lack a split for income for AI licensing within subsidiary rights schedules.
4. The settlement has been made to defend the ‘purchase once’ then AI training is Fair Use argument, a win for Big Tech.
5. There is a cost involved in registering with the US Copyright office, and retrospective registration, currently, will not impact this settlement, unless the scope of the settlement changes.
6. We are in touch with the UK & Australian Society of Authors to understand what this means for international authors and publishers, and legal teams are active.
7. A meeting on September 25 is the next milestone, and we will update our members.
8. From the information received, the authors and publishers who will see some of the settlement money are authors who have had US editions published, and registered with the Copyright Office at or around the time of publication, and that were included in the dataset.
9. The settlement has been rejected as being too narrow in scope.

US Copyright Office

What happens next: https://authorsguild.org/news/what-authors-need-to-know-about-theanthropic-settlement/

Is your book involved? This is the Lib Gen dataset:  https://www.theatlantic.com/technology/archive/2025/03/search-libgendata-set/682094/