Bartz v Anthropic Settlement Gets Preliminary Approval – Key Takeaways

On Thursday, Judge Alsup of the Northern District of California granted preliminary approval of a settlement in the class action lawsuit brought against Anthropic by three book authors whose books were among the several million downloaded by Anthropic. Annelise Levy at Bloomberg Law has a good write-up of the hearing where Judge Alsup indicated his approval. 

I won’t repeat the entire case history here — it has had numerous twists and turns, which you can read about in our previous posts. 

One important thing I will remind you of is that the case has significantly narrowed in scope — the court already decided earlier this year that Anthropic’s LLM training was fair use, and that it was permissible to buy books and digitize them for future use.  The only issue that remained was whether it infringed copyright by downloading and storing copies of books downloaded from sites like LibGen. That was slated to go to trial when the parties settled — for what we now know are rights associated with 482,460 books that fall within the certified class. In exchange for a release from liability for use of those books (note that this is broader than the pirating claim that was to be litigated, though only backward looking, covering past conduct), Anthropic will pay $1.5 billion. 

The last time I wrote about this case, it was up for approval for the first time before Judge Alsup, which he declined to grant until the parties resolved several issues, chief among them how to handle notice to rightsholders, and how to handle potential intra-class conflict about ownership of rights and payout of settlement proceeds. 

The docket for this case is now quite confusing — you’ll find many filings with proposed forms and notices that the Class Counsel proposes the court adopt. Class Counsel will eventually post a simplified version of the case history on the class action website, https://www.anthropiccopyrightsettlement.com/. This is also where you will be able to search to see if your book is included. 

For now, some key documents to focus on that are responsive to the changes and additional details Judge Alsup asked for are here: 

• This document, in which both the Class Counsel and Anthropic respond to Judge Alsup’s (whopping) 34 separate questions about how exactly the settlement notice, payout, and administration would happen. 
• This supplemental brief from Class Counsel, making the case for the court to grant preliminary approval, in which Class Counsel explains some of the new developments since the settlement was first proposed. 
• This “Plan of Allocation” explains how proceeds from the settlement will be divided up amongst rightsholders and beneficiaries.
• This Amended Proposed Order , which is what Class Counsel has asked the court to adopt, gives the parties the go-ahead to implement the plans in the proposed settlement.  

The basic structure of the settlement is the same. Here’s what we learned in the new plan that Jusge Alsup approved: 

Finding and notifying rightsholders

While 482,460 books that are the subject of the settlement sounds like a lot less than the much larger 7 million that Anthropic allegedly downloaded, it’s still a large number to sort through for finding and notifying rightsholders.  

Figuring out how to notify class members is an important step in any class action. As far as notice goes, Class Counsel states some pretty interesting numbers:  “The updated Class List contains email or mailing addresses for approximately 97% of all publishers of works on the Works List (amounting to a publisher address for 99.4% of listed works) and approximately 66% of all authors of works on the Works List.”

A representative for JND (the named settlement administrator) filed an affidavit giving a few more details: according to it, they worked to compile a complete list of unique authors and publishers for each work on the Work List. This amounted to 367,824 unique authors, and 16.237 unique publishers. JND went on to state that is has found contact information for 243,397 unique authors (66.2% of unique authors) accounting for 392,707 (81.3%) of books on the Works List, and 15,786 unique publishers (97.2% of the publishers). The same affidavit reported that over 49,000 people have submitted contact information through the website Class Counsel established for the settlement. 

Those numbers are actually pretty impressive, though they come with some caveats. For instance, some contact information is derived from copyright registration records. That’s a great place to start but often is outdated and no longer accurate. There are other minor quibbles (e.g., no concerted effort to search for other types of righthsholders such as estates, institutional owners),  that make one think that the contact information identified may have some flaws, but overall it seems like a good start. 

Finally, it’s worth observing that it’s not at all surprising that there is far better contact information for publishers than there is for authors, which reinforces something I’ve been concerned about since this settlement was announced: that publishers will end up faring far better than most authors. Publishers are more likely to be aware of the suit, will be more organized in pursuing claims, they have more resources to make claims about payouts and ownership rights, and ultimately they will receive more money in the end. 

Dividing up settlement payouts

One issue I (and apparently Judge Alsup, who asked lots of questions about this) have been keenly interested in is how the settlement with divide settlement dollars when one work has multiple claimants, especially when those claimants disagree about payouts or even whether to stay in the settlement at all rather than opt out. 

To start, the “Plan of Allocation” spells out the basics pretty cleary: in cases of multiple claimants fo the same work, the “the non-mandatory, default author(s) share and publisher(s) share are each 50% (the “Default Option”).” 

The Plan makes clear that this isn’t mandatory— “every claimant will have the option to deviate from the default rules set forth above by checking a box on the Claim Form and submitting relevant information and/or documentation (including publishing agreements) that govern their entitlement to the proceeds of the settlement.” It is important to note that for non-education works, documentary proof must be submitted in order to deviate from this 50/50 default. Given how poorly many authors and publishers manage their publishing contracts, and given how authors often lack the legal resources to challenge their publishers, I’m guessing that more often than not the 50/50 split will prevail, or if an alternative is proposed it will likely favor the publisher. 

It’s also worth observing that the 50/50 default split means that in cases of co-authors, the “author” share will be split among them, while the publisher takes the entirety of the other 50%. The Plan gives this example to illustrate: “For example, in the case of a book published by a university press with three co-authors, all four parties (the university press and the three co-authors) submit valid claims, and all four parties elect to follow the default rules. If the per-work allocation is $3,000, then $1,500 is distributed to the university press and $500 is distributed to each co-author.”

There are a few exceptions to this “default” rule. One is that it doesn’t apply to “Education Works” which the Plan of Allocation describes as “works published by education publishers. Education publishers specialize in creating and publishing works,including but not limited to textbooks, for the instruction of students and professionals, which are distributed for and through educational and professional markets. The searchable directory of works accessible on https://www.anthropiccopyrightsettlement.com/ will identify which works are Education Works.” For these education works, rightsholders and beneficiaries can make claims to a non-50/50 split without documentary support. It is only when “the claimants to an education work submit conflicting responses, [that] the Settlement Administrator will request further information and/or documentation.” 

This makes some practical sense as publishing contracts for textbooks can vary quite a bit. In some cases they are treated as works for hire. In many cases splits for payouts for things like litigation can deviate wildly. Interestingly, the Plan just requires the claimant’s  “good faith” assertion of the applicable contractual relationship. Perhaps an acknowledgement that often these contracts are difficult to dig up (e.g., this suit where education publishers had multiple copyright infringement claims dismissed for failure to provide any evidence that they had publishing contracts with authors). But, the reality here may be that many be that if one side or the other does not show up to dispute a claim, payout will be resolved in favor of a claimant who provides no actual documentation of their right to the funds.

The second big exception is when one is a sole owner, where the sole owner gets 100%. This could be the case for, for example, a self-published author, or where rights have reverted to the author.  But, it also covers many works such as works owned entirely by the publisher because it was commissioned as a work-made-for-hire, or where the publisher owes authors no royalties. I think the latter will be particularly common for multi-chapter scholarly books, where publishers typically get a grant of rights (which could be as broad as a grant of ”all right and interest in copyright… in all formats and through any channels now known or later developed,” for example ) and rarely pay authors royalties on sales. 

Finally, how money is paid out can make a difference in ultimately what one receives.  The settlement does include one feature that helps to protect righthodlers of books with multiple people with an interest in the copyright: it provides that if one co-owner (e.g., the publisher) makes a claim, the settlement administrator will automatically send a notice to any known co-owners (e.g., the author), and will by default send that person a check. However, if the check isn’t cashed within 18 months, the money will revert back to the claimant who did file a claim. This is probably better than the alternative, but I do worry 1) that sending checks of potentially thousands of dollars to unverified addresses could lead to some fraud, and 2) that authors who are unaware, ignore the letter (since they are not expecting it), etc. will not cash those checks and thus lose out entirely. 

Disputes among class members

This part of the settlement is straightforward, though disputes could come in a few forms. One is about inclusion in the settlement at all. Class members have the opportunity to opt out within “104 days after Preliminary Approval (i.e., 45 days after the Notice End Date)”, and as is made clear in the 34 question responses, if one class member with an interest in a book opts out, the entire work is opted out and any co-authors or the publisher will no longer be included in the settlement for purposes of that work. Any claims related to that book are not covered by the settlement, meaning the author or publisher could now independently pursue a separate copyright claim against Anthropic, seeking damages or asking Anthropic to not retain or train on their books. 

When all co-authors and publisher want to remain in the settlement, but have a dispute over who gets how much, the solution is to send the dispute to a “Special Master,” which is essentially a judge who presides over special aspects of the settlement, though it is different from a normal judicial setting in that the results would be binding, and there would be no right to appeal. 

Conclusion

There is lots more to say about this. Unfortunately, the Works List that forms the backbone of this settlement will not be fully released, making it hard to do a more thorough analysis of what likely outcomes might result for different categories of works. There will be a searchable database coming on the settlement website, though, so be sure to check it out, especially if you think your book may be included. We plan to run searches and, to the extent possible, inform Authors Alliance members of potential matches for their works in the list.