Image – Getty: Sittipol Sukuna
By Porter Anderson, Editor-in-Chief | @Porter_Anderson
See also:
Interview: Niels Famaey on Fighting Piracy in Belgium
UK Publishers and Cambridge Call Out Meta and Piracy in Generative AI Training
French Publishers Prevail Against Manga Pirate Site Japscan
Copyright: Publishers Join Creative Industries’ Protest of Europe’s AI Act Implementation
Maria A. Pallante: ‘A Very Positive Development’
Many of our international readers may not have been aware of the class-action lawsuit against the San Francisco-based AI company Anthropic. A new inflection point in the case this week has provided a good moment to acquaint yourself with this one, which some observers feel has the potential to have substantive effect in a dimension of large language model training on unlicensed copyrighted content—at least when that content is sourced from pirate sites.
So large is the class being formed—of both publishers and authors—that Anthropic in its defense is asserting that paying damages to publishers and authors could bankrupt the AI company.
That’s something Beatrice Nolan covered in her July 28 article for Fortune on Tuesday (August 12). She wrote, “A class-action lawsuit against Anthropic could expose the AI company to billions in copyright damages over its alleged use of pirated books from shadow libraries like LibGen and PiLiMi to train its models. … Legal experts warn that statutory damages could be severe, with estimates ranging from US$1 billion to more than US$100 billion.”
“The potential scope of works that Anthropic allegedly infringed numbers in the hundreds of thousands to millions and consists of works from thousands of copyright owners, both authors and publishers.”Maria A. Pallante, AAP
And that, in essence is why this case, building for a year, is drawing so much focus and response this week as a tight September 1 deadline looms, per the court, for the plaintiffs’ counsel to deliver a list of works that are material in the case. The potential length of that list is both daunting to those working to create it and exhilarating to many who have hoped for what James Darley at Technology Magazine (August 12) has suggested could be “a moment of reckoning” for companies using unlicensed copyrighted content to train LLMs, even with the pertinent element in the judge’s purview being the use of pirated sites to procure that content.
Anthropic was founded in 2021 by former employees of OpenAI (the producer of ChatGPT). The company’s main product is a suite of large language models (LLMs) called Claude. Perhaps not without some evident irony in light of current allegations, the company’s self-styled emphasis on ethical and safe development in artificial intelligence stands out when you peruse its site.
Almost a year ago, on August 19, 2024, authors filed a lawsuit against Anthropic for allegedly training its models on copyrighted works. At least some of those works are asserted to have been downloaded from pirate book sites, such as Library Genesis (LibGen). That lawsuit, Bartz v. Anthropic, was filed by authors, who charge that Anthropic has made copies of the text of something around seven million books from pirate sites Pirate Library Mirror (PiLiMi), a mirror site of Z-Library, and LibGen.
In a somewhat curious opinion issued on June 23, Judge William Haskell Alsup of the District Court for the Northern District of California said that copying the texts of the copyrighted books without payment or permission was protected as “fair use,” because such usage was “transformative”—a criterion that indicates a new work has substantially altered an original work’s character and purpose.
At the same time, Judge Alsup said that the attainment of such books from pirate sites was not covered by fair use principles.
Late on Monday (August 11), the Association of American Publishers‘ (AAP) president and CEO Maria A. Pallante issued a memo clarifying that on July 17, Judge Alsup had also “certified a class that includes publishers as well as authors, to accommodate rights ownership questions.” A trial date set for December 1, she notes, “requires an accelerated notice schedule to the class, requiring members of the class to understand their rights and options.”
What Pallante is doing, then, is alerting publishers to the fact that they may have a direct interest in this case: “The potential scope of works that Anthropic allegedly infringed numbers in the hundreds of thousands to millions,” Pallante writes to publishers, “and consists of works from thousands of copyright owners, both authors and publishers. Your house and your authors may each be part of the class.”
Defining the Class for Publishers
As defined by Judge Alsup—as Pallante relays it to the publishers—the class includes:
- The beneficial or legal owners of the exclusive right to reproduce copies of any book downloaded by Anthropic (1) through Library Genesis (LibGen) in July 2021, or (2) through Pirate Library Mirror (PiLiMi) in July 2022; and
- “Books” here means any work (1) possessing an ISBN or ASIN, that was (2) registered with the US Copyright Office (a) within five years of the work’s publication and (b) prior to being downloaded by Anthropic or within three months of publication.
The Update for Publishers
The key information that Pallante is getting across to publishers is that the group of law firms working to support publishers and authors is growing—in what may be an enormous class that has to be ready very quickly.
Maria A. Pallante
“No publisher or author will have to pay any out-of-pocket fees or expenses to the firms,” Pallante points out.
On Monday, she writes, “The class action law firm Edelson PC and the copyright firm Oppenheim + Zebrak, LLP (O+Z) filed a Notice of Association of Additional Counsel to enter the case as additional plaintiffs’ counsel (“Publishers’ Coordination Counsel”), joining the original plaintiffs’ firms of Susman Godfrey LLP; Lieff, Cabraser, Heimann & Bernstein LLP; and Cowan DeBaets Abrahams & Sheppard LLP.”
Speaking for the AAP leadership, she writes, “We believe the additional counsel is a very positive development that will help to ensure that the class interests are optimally represented without conflict.
“Specifically, the Publishers’ Coordination Counsel will provide the publishers’ perspective and assist with trial preparation and strategy, as well as the trial, assembling the class list, class notice, and settlement discussions.”
AAP’s team itself is to serve as point of contact for the Publishers’ Coordination Counsel, as well as for publisher-members of the class. Some of the biggest publishers in the association are also in direct touch with the law firms, and the association can connect other publishers interested in direct contact.
“To be clear,” Pallante writes, “this suit is a class action and the class counsel for the plaintiffs therefore have a duty to all authors and publishers in the class, with the Publishers’ Coordination Counsel representing the interests of publishers in the common goal of maximizing the per-work recovery for the class.
“Based on the court’s class definition,” she writes, “counsel for both sides are now working to provide the court with a draft notice form and a list of works at issue, due to the court August 15 and September 1, respectively. Once approved by the court, notice forms will be distributed widely to class members, including US publishers. Each class member so notified will then have to decide whether to stay in the litigation by following the court’s instructions.
“There is no decision for you to make now,” Pallante’s memo to publishers goes on. “The court will issue a notice that will provide you with additional information, including your options with respect to this litigation—primarily, whether to stay in the class or exclude yourself. We’ve spoken to the Publishers’ Coordination Counsel about holding a town hall at the time notice goes out so that they can explain the process, go into more detail about the options you have, and answer any questions. We do not know yet when notice will go out and accordingly do not know when the town hall will be, but we will let everyone know well in advance.”
The Association of American Publishers, she adds, while unable to provide legal advice to a publisher, can speak with houses about the meaning of each option. “For publishing houses that remain in the class, we will ensure that you have clear technical and procedural information and that you are well represented.”
An Additional Note About Anthropic
In a move unrelated to the class-action lawsuit discussed in this article, Anthropic on Tuesday (August 12) was reported by Reuters to be offering its Claude AI model(s) to the United States government for nominal fee. OpenAI announced a similar offer last week with ChatGPT enterprise at a rate of US$1 per federal agency for the next year, the report has it.
As Ashley Capoot describes the move at CNBC (August 12), “It’s the latest example of how major artificial intelligence companies have been looking to deepen their ties to policymakers and regulators in recent months. … [In July,] the US Department of Defense announced contract awards of up to US$200 million for AI development at Anthropic, Google, OpenAI and Elon Musk’s xAI.”
A Programming Note
Image: FBM, Holger Menzel
The 2025 Frankfurt Rights Meeting program ahead of Frankfurter Buchmesse (October 15 to 19) opens on September 2 with The AI Dilemma: Balancing Risks and Opportunities. The session features HarperCollins’ Chantal Restivo-Alessi and the FALA Agency’s Fatimah Abbas. An article on the full Rights Meeting series of pre-Frankfurt events is here, with dates, timing, pricing, and other details. The program comprises four weekly digital events and an in-person reception at Frankfurt in October.
More from Publishing Perspectives on artificial intelligence is here; more on copyright and international book publishing is here; more on the United States’ market is here, and more on the Association of American Publishers is here.
Wherever our readers are in the world, they use our free daily email to be sure they don’t miss a story. Sign up now.