Claude AI to pay US $1.5B to settle copyright class action

Several authors commenced class action lawsuit against Claude AI (Anthropic) in the San Francisco District Court. The authors allege Anthropic infringed their copyright by using millions of pirated and purchased books to train its artificial intelligence (AI) chatbot, Claude AI. Anthropic has agreed to settle the class action approximately $1.5 billion US dollars, avoiding further trials. However, Judge Alsup has since postponed approval of this settlement due to its misleading nature and concerns surrounding a ‘forced’ deal.^[1] In its initial judgments, prior to the settlement being raised, the Court case interpreted the meaning of ‘fair use’ in the United States, particularly regarding its ambiguity in the age of evolving technology and AI systems.

Background of the Anthropic case

The class action complaint and order on fair use documents filed in the San Francisco division of the United States District Court outlines the details of the Anthropic case. Anthropic is an AI software firm which allegedly ‘stole’ millions of books from several authors to train Claude AI.

Anthropic took and reproduced millions of books, which had been both pirated from the internet and purchased. Purchased books were digitised, with the paper copies being discarded. These were then added to a digital ‘research library’ for the purpose of training its large language models.^[2] No permission or license was sought to use these materials in training these models. Further, this was not an isolated incident but rather occurred continuously over the course of several years.^[3]

In August 2024, three (3) authors brought an action against Anthropic alleging that their copyright had been infringed through Anthropic downloading pirated copies and reproducing them to train their AI models.[4]

What constitutes ‘fair use’ in Australia?

In Australia, ‘fair use’ is instead referred to as ‘fair dealing’, though similar principles underpin them. Section 40 of the Copyright Act 1968 (Cth) states:

“a fair dealing with a literary, dramatic, musical or artistic work, or with an adaptation of a literary, dramatic or musical work, for the purpose of research or study does not constitute an infringement of the copyright in the work“.^[5]

How had the Anthropic class action approached copyright infringement?

In Anthropic, the company claimed that using the authors’ works for training AI was “quintessential fair use“. Anthropic indicated that ‘fair use’ existed because:

they were studying the plaintiffs’ writing;
the information being extracted was uncopyrightable;
the information being obtained was being used to create ‘revolutionary technology’; and
obtaining the information did not deprive the plaintiffs of a market for their books.^[6]

The Court conducted an analysis of section section 107 of the United States’ Copyright Act, considering:

“the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
the nature of the copyrighted work;
the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
the effect of the use upon the potential market for or value of the copyrighted work“.^[7]

Following a consideration of all factors it was determined that the copies of books used to train specific large language models, and those used to convert purchased print library copies into digital library copies, were justified through fair use.^[8] This is because they favoured the development of transformative technology.^[9]

However, the downloaded pirated copies of books used to build a central library were not justified by fair use because the copies were going to be ‘retained forever’ for a ‘general purpose’, which was outside the scope of training AI models.^[10] This same principle applied to any copies downloaded but not used for training, indicating that fair use will only apply where works have been taken for that use alone.^[11]

Settlement delay

As a result of this infringement, Anthropic has agreed to settle for approximately $1.5 billion US dollars.^[12] However, Judge Alsup has postponed approval of this settlement because many details of the case have not yet been decided on.^[13] For example, the number of works included in the settlement is not yet complete, and it is important that this is resolved before settlement is agreed upon to avoid further disputes. Judge Alsup has provided Anthropic with a deadline of 15 September to submit a list of works, and if this is unsatisfactory, the case will be forced to go to trial on 1 December 2025.^[14] The case itself teaches that companies should exercise caution when reproducing works for the purposes of training AI and large language model.

Links and further references

Legislation

Copyright Act 1968 (Cth)

US Copyright Act

Further information about laws relating to the adoption of artificial intelligence

If your business needs advice on adoption of artificial intelligence and copyright law, please contact us for a confidential and obligation-free discussion:

Malcolm Burrows B.Bus.,MBA.,LL.B.,LL.M.,MQLS.
Legal Practice Director
T: +61 7 3221 0013 (preferred)
M: +61 419 726 535
E: mburrows@dundaslawyers.com.au

Disclaimer

This article contains general commentary only. You should not rely on the commentary as legal advice. Specific legal advice should be obtained to ascertain how the law applies to your particular circumstances.

[1] Order re hearing on motion for preliminary approval of settlement.

[2] ECF-231-Order-on-Fair-Use.pdf.

[3] ECF-231-Order-on-Fair-Use.pdf.

[4] ECF-231-Order-on-Fair-Use.pdf p 8.

[5] Copyright Act 1968 (Cth) s 40.

[6] Anthropic’s Motion for Summary Judgment.

[7] US Copyright Act s 107.

[8] ECF-231-Order-on-Fair-Use.pdf p 30.

[9] ECF-231-Order-on-Fair-Use.pdf p 30.

[10] ECF-231-Order-on-Fair-Use.pdf p 31.

[11] ECF-231-Order-on-Fair-Use.pdf p 31.

[12] Motion for Settlement (September 2025).