Anthropic's $1.5 billion copyright settlement with authors could set precedent in AI industry
The settlement arises from Bartz et al v Anthropic PBC, a case filed in the Northern District of California, by authors including Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson. The authors alleged that Anthropic (founded by former OpenAI employees) had downloaded hundreds of thousands of digital books, including pirated copies, from online shadow libraries such as Library Genesis and Pirate Library Mirror to train its Claude family of large language models ("LLMs").
Anthropic had also purchased copyright-protected books, torn off the bindings, scanned each page and stored them in digitised, searchable files. Anthropic's aim was to amass a central library of "all the books in the world" to retain "forever". The authors sought damages, injunctive relief, and accountability for the unauthorised acquisition and use of their intellectual property.
Anthropic defended the claim by asserting that its actions constituted fair use and were necessary for advancements in AI development.
On 23 June 2025, the court ruled as follows:
- Granted summary judgment in relation to Anthropic's use of copies of the authors' works to train LLMs, finding the use "quintessentially transformative," and thus fair, as the models were trained not to replicate or supplant the works, but "to turn a hard corner and create something different". The AI models used the books to learn patterns and generate new text, but did not reproduce or distribute the original works;
- Granted summary judgment in relation to Anthropic's conversions of lawfully purchased print copies to digital use, which the court also described as "transformative", noting that the format change "added no new copies, eased storage and enabled searchability" and was therefore fair use; and
- Denied summary judgment in relation to Anthropic's copying of pirated works for use in the central library, ruling that such copying was not fair use. Anthropic employees had said copies of works, including pirated ones, would be retained "forever" for "general purpose", even after Anthropic determined they would not be used to train LLMs. This was not fair use. Judge William Alsup stated that, "…. the person who copies the textbook from a pirate site has infringed already, full stop. This order further rejects Anthropic's assumption that the use of the copies for a central library can be excused as fair use merely because some will eventually be used to train LLMs."
Don't miss a thing, subscribe today!
Stay up to date by subscribing to the latest Intellectual Property insights from the experts at Fieldfisher.
Subscribe nowThe court scheduled a trial for Anthropic's potential liability in relation to the pirated books for December 2025, however, the parties reached a settlement before the trial. On 25 September 2025, Judge Alsup gave his preliminary approval for the proposed 'historic' settlement:
- Anthropic to pay damages of $1.5 billion for its past use of pirated books representing $3000 for each of the approximately 500,000 copyrighted works alleged to have been downloaded from the relevant libraries and infringed by Anthropic.
- Anthropic to destroy the two libraries containing the pirated works, as well as any derivative copies originating from those sources. Anthropic must confirm in writing that the destruction has been completed and that the allegedly infringing materials have been permanently removed from Anthropic's systems.
The settlement does not shield Anthropic from future lawsuits by authors whose works do not form part of this litigation.
Anthropic has not admitted liability and Judge Alsup has postponed the final approval hearing to April 2026 while the workability of the proposed settlement is considered.
In the meantime, authors who suspect their works may have been infringed should search the Works List, which was published on 2 October 2025, to identify whether there has been unauthorised use of their works by Anthropic. They then have until 23 March 2026 to submit their claims. The claim form offers an opportunity for authors to 'opt out' of the settlement which means rejecting any award and preserving any claims against Anthropic. The opt out window runs until 7 January 2025.
Comment
If no major objections arise, this will be one of the most significant copyright settlements in the AI era. It is thought to be the largest publicly reported copyright recovery in US history. This is a significant step in the AI space and will set a precedent for how courts may treat AI training on pirated content. For those in the US, this is also a reminder to rightsholders to ensure that any copyright work is registered (not required in the UK) because only registered works are eligible for compensation.
The settlement demonstrates that the intellectual property in creative works is both increasingly valuable and increasingly at risk. Those who understand that genuine technological advancement must be matched with responsible, transparent management of intellectual property rights will be best placed to minimise legal risks and fully capitalise on the value of IP assets.
In addition to Bartz et al v Anthropic, Concord Music and other music publishers have filed a separate claim against Anthropic, focusing on the alleged unauthorised use of copyright-protected song lyrics in AI training data. In an interesting turn of events, following Anthropic's $1.5 billion settlement with the authors, the music publishers attempted to add piracy claims to their lawsuit. However, the court denied their request, ruling that they had failed to investigate the piracy claims earlier and adding it at this late stage would unfairly delay the case. Trial is expected to take place early next year.
This action adds to the growing scrutiny of how AI developers source and use creative works, and as courts and rightsholders continue to test the boundaries of fair use and copyright in the context of AI, these cases are will continue to set important precedents for the industry going forward.
With special thanks to Solicitor Apprentice Jessica Gurzynski for co-authoring this article.
