Landmark US Court Decision: AI Training on Books Recognised as Fair Use - What This Means for the Industry

30.06.2025

• Background
• Court's Decision: What Constitutes Permissible Use
• What Is Transformative Use?
• What Lies Ahead for Anthropic
• Criticism from the Authors Guild of America
• Implications for Other AI Companies
• Practical Recommendations for Business
• Contact our lawyer for more details

On 23 June 2025, the US District Court for the Northern District of California issued a landmark summary judgement in Bartz v. Anthropic, concerning the use of books for training generative AI models. Judge William Alsup partially granted Anthropic's motion to dismiss, recognising part of the company's actions as "fair use," whilst leaving certain allegations intact.

Background

Anthropic utilised a vast corpora of texts to train its AI model Claude. Specifically, the company digitised millions of purchased books and used downloaded copies to create a "central library," which was subsequently employed for model training. The question before the court was whether such use of books could be considered permissible under copyright law.

Court's Decision: What Constitutes Permissible Use

The court identified three distinct categories of use:

• Copies for LLM training (recognised as transformative)
• Conversion of lawfully purchased print books into digital format (fair use as format-shifting)
• Creation of a general library from pirated materials (NOT fair use)

Thus, the court noted that digitisation of relevant books and subsequent use for training the Claude model constitutes transformative use. This means the books were not merely copied, but were utilised for analysis and creation of new texts, which qualifies as a transformative purpose and is supported by the fair use doctrine.

However, this principle applies only when using legitimate, rather than pirated, content.

Importantly, the court noted that no infringing content reached Claude users thanks to filtering systems, which became a key factor in the decision.

What Is Transformative Use?

Transformative use occurs when a work is used not for the purpose of reproduction, but to create something new that expands public knowledge or serves another beneficial purpose. In this context, Anthropic used texts to train its AI model in such a manner that the result of its work differs substantially from the original works. This renders the use of such materials transformative and, consequently, permissible under fair use.

Judge Alsup also cited the case of Authors Guild v. Google Inc., where Google digitised books and provided them in textual and image formats, which the court recognised as transformative use.

It is important to note that such actions do not harm rightsholders' interests, as the end product (searching for information about books, rather than their complete reproduction) was aimed at creating new functionality.

What Lies Ahead for Anthropic

The court rejected the fair use defence solely for the pirated library, leaving this portion of the allegations for further proceedings. The case will continue specifically regarding the use of illegally obtained copies.

Criticism from the Authors Guild of America

The Authors Guild of America is the largest and one of the most influential organisations representing writers' and authors' interests in the United States. It expressed concern regarding the court's decision in Bartz v. Anthropic, which concerns the use of pirated books for training AI models. In its statement, the Guild noted that whilst the court recognised Anthropic's piracy activities as unlawful, the decision that using pirated and digitised books for AI training constitutes fair use is erroneous.

The Guild expressed the following principal objections:

• The Guild disagrees with the recognition that using pirated and digitised books for AI training constitutes fair use, considering this a violation of copyright.
• The Guild criticises the court for ignoring potential harm to authors, including market loss and incorrect attribution of their works by AI.

The Guild believes the decision should be overturned as it contradicts established judicial precedent.

Implications for Other AI Companies

This decision is significant not only for Anthropic but for the entire AI industry. The court recognised that using copyrighted materials for AI training may be lawful based on the fair use concept, provided materials are used to create new information or technology, rather than for direct copying and distribution.

It is important to note that the court emphasised such actions must be transformative and must not interfere with rightsholders' ability to use their works for other purposes. For example, training AI on texts, as in Claude's case, was deemed permissible because the end product (AI generating new text) was substantially different from the original works.

Practical Recommendations for Business

If your company uses copyrighted materials for AI training, it is important to demonstrate that use of these materials is transformative. This may include using works to create new products or improve existing technologies, rather than merely copying them.

Despite the court recognising transformative use of digitised books, it is important to conclude appropriate licensing agreements or purchase content from rightsholders to avoid legal risks and enhance transparency in using copyrighted materials.

The Arbitration & IT Disputes team is prepared to assist you in resolving matters related to copyright compliance and developing appropriate licences for data use, as well as in preparing strategies to minimise risks in the area of using copyrighted materials for AI training.

Authors: Tserakhau Kamal, Vainilovich Krystsina, Shumilau Hleb

---

Attention Journalists: Use of REVERA website materials in publications is only allowed with our written permission.