Anthropic has agreed to pay $1.5 billion to settle charges that it illegally downloaded hundreds of thousands of books from pirate databases Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi) as part of its effort to train its large language models. The AI company has also committed to destroying all the pirated copies of books which it had stored in the training process.
The agreement brings a swift resolution to a case that only took root this July, when U.S. District Judge William Alsup ruled that a lawsuit brought by three authors—Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson—challenging the legality of Anthropic’s use of books from pirate websites could move forward as a class action. At the same time, the judge also ruled that Anthropic’s overall training of its Claude LLMs on authors’ works was protected by fair use.
Attorneys for the authors called the agreement the largest publicly reported payout in the history of U.S. copyright litigation. “This landmark settlement far surpasses any other known copyright recovery,” said Justin Nelson of Susman Godfrey and co-lead counsel for the plaintiffs in a statement. “It is the first of its kind in the AI era, and sets a precedent requiring AI companies to pay copyright owners.”
Representatives from the Authors Guild and the Association of American Publishers also hope that the agreement will prompt AI companies to follow copyright law in training their LLMs. “The proposed settlement provides enormous value in sending the message that artificial intelligence companies cannot unlawfully acquire content from shadow libraries or other pirate sources as the building blocks for their models,” said Maria Pallante, president and CEO of the AAP, in a written statement.
Authors Guild CEO Mary Rasenberger echoed Pallante’s sentiment, saying the agreement is “an excellent result for authors, publishers, and rightsholders generally, sending a strong message to the AI industry that there are serious consequences when they pirate authors’ works to train their AI, robbing those least able to afford it.”
As the agreement now stands, some 500,000 works are covered by the settlement, with each work eligible to receive about $3,000. (A licensing deal reached by HarperCollins last November with an unnamed tech widely believed to be Microsoft provides that the author and publisher would split a $5,000 per book fee).
Although the lawsuit charged that Anthropic had copied as many as seven million titles, the settlement pertains only to works that meet certain criteria established by the court. According to the agreement, to be part of the settlement, the work must have either an ISBN or ASIN and must have been registered with the U.S. Copyright Office within three months of publication, or within five years of publication and before LibGen and PiLiMi began downloading titles, in June 2021 and July 2022, respectively. The discrepancy between the works covered by the settlement and the number of books believed to be pirated is due in part to the fact many files were either not books or duplicates, as well as because a large number of titles were not registered with the Copyright Office.
A hearing is set for September 8 for the agreement to receive preliminary approval from Judge Alsup. If, as expected, Judge Alsup approves the agreement, final approval is expected to take place sometime next year. Until that last approval, there are still many issues to be resolved.
For starters, the parties are working on creating a final list of the works included in the agreement, which is due to the court by October 10. From that list, a searchable database will be created where authors and publishers can search for their works. If the final number of works exceed 500,000, Anthropic will add $3,000 for each additional title. (Authors who think their work may have been included in the pirated databases used by Anthropic can learn more about how to file a claim for compensation at AnthropicCopyrightLawsuit.com.)
If the settlement wins preliminary approval today, an administrator is expected to send notices out this fall, which will instruct class members how to file a claim and provide other information about their choices. Authors will have the option to opt out of the agreement and could file their own lawsuit, though that would be an expensive undertaking. The settlement only covers “past claims” and does not cover any claims of conduct against Anthropic that postdates August 25, 2025, the date of the binding agreement.
The Authors Guild and AAP were not part of the lawsuit, but did provide advice. Since both publishers and authors are eligible to receive a payout from the fund, attorneys representing the class have proposed establishing an Author-Publisher Working Group (APWG), which will work to determine the fair distribution of funds to class members. Under the proposal, Rasenberger and Pallante will provide any recommendations to the court by Sept. 30.
The final approval hearing will be held after any comments or objections are received after which the court will determine whether to grant the settlement final approval. According to the court order, once the settlement is approved, the settlement administrator will start sending the initial tranche of the settlement fund payments within a month.