NXIP advises author in $1.5 Billion Anthropic Class Action Settlement

Bartz et al. v Anthropic PBC.

 Where we are:

“A settlement has been reached in a class action lawsuit claiming that Anthropic infringed protected copyrights by downloading books in two allegedly pirated online datasets called Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi).  Anthropic denies all allegations and argues that its use of the downloaded datasets was fair use”.

Background to the Class Action:

The alleged infringement took the form of Anthropic constructing their AI system, a large language model (LLM) system called Claude, which they then commercialised to generate significant revenue. LLMs are AI systems trained on long-form text datasets to construct models that are capable of statistically analysing and imitating a human's capacity to connect concepts and words into sensible utterances.  

The Claude system is alleged to have been constructed using long-form text datasets including The Pile, which in turn incorporates a dataset called "Books3".  The creator of "Books3" apparently created it by downloading a compilation of 196,640 books (at least some of which are still protected under copyright law) called Bibliothik. Bibliothik is referenced in many places as pirated content.

The Plaintiffs asserted that this use of copyright-protected books from Bibliothik to construct the Claude AI system did not constitute fair use, and thus infringed the rights of the copyright holders. The Defendants admitted no liability but proposed to enter into a no-fault settlement with the Plaintiff class.

The Settlement:

A class member is a legal or beneficial owner of work that was included in the LibGen or PiLiMi datasets downloaded by Anthropic; has an International Standard Book Number (ISBN) or Amazon Standard Identification Number (ASIN); and was registered with the U.S. Copyright Office within five years of first publication.   

With these factors applying, then class members have until March 30, 2026 to file a claim to request to be eligible for payment from the $1.5 Billion Settlement Fund.

The court has scheduled a hearing on April 23rd in San Francisco to consider whether to approve the Settlement.

Role of NXIP

After receiving notification of being an author of works named in the class action, we were approached by the author concerned that a number of their works had been digested and used to train Claude. By assessing the circumstances of their publications and the author’s contractual relationship with the publisher we were able to advise and assist the author in submission of their claim.

A searchable database of all works covered by the Settlement can be found at www.AnthropicCopyrightSettlement.com/lookup.

Practice Points

Whilst watching movies or buying CDs and records as a child it was impossible not to notice the “copyright infringement is theft” notice displayed prominently on the packaging or start of a movie.  A statement which has probably inspired more than one child to wonder what that's all about!

Then, during computer science at university, the mantra “Garbage in, Garbage Out”, was drilled into us students to realise that accurate, validated results and unbiased data are crucial for reliable outputs.

We seem to be rapidly converging at a point with Generative AI where the economic scale of some undertakings can bypass and ignore intellectual property rights and where “Garbage in” has somewhere along the way gone from being a warning to being turned into a workflow. 

In our UK patent attorney universe we see many patent attorneys with very limited understanding of AI models considering “which one to choose”, based on user friendliness and cost, whilst ignoring the realities of factors such as the origin of training data and confidentiality.

The training data may originate from pirated works and so by using an AI chatbot, the attorney not only contributes to an industry that is avoiding offering fair reward to creative authors, but has no visibility of the accuracy of that training data.  Furthermore, by poor choice of chatbot, such as a free, cloud based Chatbot, the attorney is making an exchange of their client's know how in exchange for a quick, cheap fix of questionable return.

Recent changes as of 2026 to the EPO Guidelines for examination now introduce dedicated guidance on artificial intelligence, confirming that parties remain fully responsible for AI-assisted filings and formalising AI use within EPO procedures.   This guidance dovetails with the guidance issued by EPI which can be found at the following link: epi Information | epi Guidelines: Use of Generative AI in the Work of Patent Attorneys

If you have any questions about NXIP’s use of AI or how we are helping clients assess risk and opportunities while incorporating AI into a business strategy then please contact us at:

alex.turnbull@nxip.co.uk