Anthropic settles AI book piracy lawsuit with US authors

TL;DR

• Anthropic reached a proposed class settlement with US authors over copyright claims tied to training Claude AI on pirated works. The Verge
• The settlement would allow the company to skip a December trial that could have determined damages; terms are not fully disclosed yet.
• The suit stems from allegations that Anthropic trained its models on an open‑source dataset filled with pirated material, raised by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson.
• The settlement is expected to be finalized on September 3; a lawyer representing the authors called it historic, while Anthropic declined to comment. The Verge

Context and background

Anthropic has faced a class-action copyright dispute filed by a group of US writers who accused the startup of training its Claude AI models on pirated works found in an open-source dataset. The lawsuit has been in play as the parties negotiated the terms of a potential resolution that would avoid a full trial. In June, Judge William Alsup ruled that training AI models on legally purchased books counts as fair use, a decision that provided a major procedural win for Anthropic but left open room for further litigation on other aspects of the case. The judge’s decision did not settle all questions about whether certain data used for training could be considered fair use in every circumstance. The Verge In July, Judge Alsup approved a class-action claim that accused Anthropic of violating copyright laws “by doing Napster-style downloading of millions of works.” This development set the stage for a trial originally scheduled for December, at which Anthropic could have faced substantial penalties, with Wired noting figures that could reach billions or more than $1 trillion depending on the outcome. The Verge The terms of the proposed settlement were not disclosed at the time of the filing, and the final agreement is expected to be finalized on September 3. Justin Nelson, an attorney for the authors, described the settlement as benefiting all class members, while Anthropic chose not to comment publicly. The Verge

What’s new

The primary new development is the negotiation of a “proposed class settlement” that would allow Anthropic to skip the trial that would have determined damages for alleged copyright infringement. The parties filed a legal agreement indicating intention to settle as a class, but the precise terms, coverage, and any payments remain confidential or unresolved as of the filing. This approach aims to provide a resolution for a large group of authors collectively rather than litigating each individual claim. The settlement is described as potentially benefiting all class members by avoiding the need for a protracted trial and related proceedings. The targeted date for finalization is September 3, with further details anticipated in the weeks ahead. The Verge

Why it matters (impact for developers/enterprises)

• The case underscores ongoing questions about how AI training data is sourced and used, particularly when allegations involve pirated or copyrighted works placed in datasets available for model training.
• A ruling that fair use can apply to training on certain legally acquired materials may influence how developers approach data gathering for model training, though the Augur of a broader class action indicates significant risk of litigation in parallel tracks.
• The existence of a class settlement, rather than a trial verdict, highlights the potential for settlements to resolve complex copyright disputes in the AI space without a verdict on damages, potentially affecting how future disputes are settled or litigated. The Verge

Technical details or Implementation

• The core allegations revolve around whether Anthropic trained Claude AI models on an open‑source dataset that allegedly contained pirated material. The phrase “Napster-style downloading of millions of works” was used by a judge in approving the class action, illustrating the nature of the claim as it was argued in court.
• The June ruling by Judge William Alsup stated that training AI models on legally purchased books can count as fair use, though this did not settle all copyright questions or the scope of data involved. The subsequent July ruling added a broader class-action dimension to the allegations, reinforcing the potential for widespread claims in this space. The Verge
• While the settlement’s terms remain undisclosed, the parties indicated a formal finalization date and public statements about the potential benefit to class members, signaling a path toward resolution without an extensive damages trial. The Verge

Key takeaways

• A proposed class settlement with US authors aims to resolve a high-profile copyright dispute related to AI training data. The Verge
• The agreement would allow Anthropic to avoid a December trial focused on damages, at least pending final terms. The Verge
• The case involved claims of training on pirated works and was influenced by a June fair-use ruling and a July class-action ruling. The Verge
• Finalization is expected on September 3; Anthropic did not comment on the settlement terms. The Verge

References

• https://www.theverge.com/news/766311/anthropic-class-action-ai-piracy-authors-settlement