Losing the battle but winning the war? What Kadrey and Anthropic might mean for AI and copyright law.

Before diving into the judgments and reasoning, it’s important to understand the legal concept that was at issue in both proceedings. Both Meta and Amazon sought to invoke the shield of the ‘fair use’ doctrine to defend their use of the plaintiffs’ copyrighted materials.

In the US, copyright law provides for a broad ‘fair use’ exception to copyright infringement where certain elements can be made out. These elements are listed in Section 107 of the US Copyright Act, which states:

“In determining whether the use made of a work in any particular case is a fair use the factors to be considered shall include—

(1) The purpose and character of the use, including whether such use is of a commercial nature or is for non-profit educational purposes;

(2) The nature of the copyrighted work;

(3) The amount and substantiality of the portion used in relation to the copyrighted work as a whole; and

(4) The effect of the use upon the potential market for or value of the copyrighted work.”

A group of authors brought a suit against Meta in the Northern District of California, including Sarah Silverman and Richard Kadrey, alleging that the technology giant used copies of 7.5 million pirated books and 81 million research papers to train its large language AI model, Llama.

Judge Chhabria found in favour of Meta’s argument that using the pirated works was defensible under copyright law as a ‘fair use’. The ‘fair use’ elements most in issue were the first and the fourth – whether the use was ‘transformative’, and whether the use caused market harm to the plaintiffs.

There was little argument that Meta’s use ‘transformed’ the source material entirely. Chhabria J wrote that Meta’s use for AI training had a “further purpose” and a “different character” than the original books. The work created by Llama was not a ‘substitute’ for the copyrighted material – for example, the LLM did not regurgitate ‘verbatim excerpts’. It turned the source material into something entirely new.

Regarding the fourth factor, Chhabria J ruled that the plaintiffs failed to produce any compelling evidence about market harm caused. He acknowledged while “that conclusion may be in significant tension with reality… it’s dictated by the choice the plaintiffs made… failing to present meaningful evidence on the effect of training LLMs like Llama with their books on the market for [AI-generated] books.” The plaintiffs couldn’t show that Meta’s use of their copyrighted work substantially diminished the market value of the original book.

Ultimately, the plaintiffs lost because they failed to present a compelling argument regarding the consequences of Meta’s actions for the market value. Chhabria J stated the court “[had] no choice but to grant summary judgment to Meta”, but warned that this ruling “does not stand for the proposition that Meta’s use of copyrighted materials to train language models is lawful… in many circumstances it will be illegal to copy copyright-protected works to train generative AI models without permission.” This is bad news for the authors in this case, but potentially good news overall for creatives concerned about the application of the ‘fair use’ doctrine to defeat copyright claims against AI giants.

It should also be noted that the issue of the piracy itself will be determined separately from the copyright infringement claim. Recent case Bartz v Anthropic demonstrates that the outlook on that front is much less rosy for Meta.

This suit was determined shortly before Kadrey, and also filed in the US District Court for the Northern District of California. The Amazon-owned AI company Anthropic was accused of downloading masses of pirated books to train its LLM model, Claude. As in Kadrey, Judge Alsup found this use was protected by the ‘fair use’ doctrine, writing that “the technology at issue was among the most transformative many of us will see in our lifetimes.” He determined that Anthropic had not breached the plaintiffs’ copyright because they “have not reproduced to the public a given work’s creative elements, nor even one author’s identifiable expressive style”.

However, Alsup J cleared a path for the plaintiffs to take Anthropic to trial over the piracy of their copyrighted works. He wrote that “Anthropic downloaded over seven million pirated copies of books, paid nothing, and kept these pirated copies in its library even after deciding it would not use them to train its AI (at all or ever again). Authors argue Anthropic should have paid for these pirated library copies. This order agrees.” The lowest statutory damage amount available to the plaintiffs for copyright infringement via piracy is $750 per book. With over 7 million books allegedly pirated, Anthropic potentially faces billions of dollars’ worth of court-imposed penalties. A date for this separate trial has not yet been set.

The relationship between the two trials creates an elliptical situation where Anthropic’s use of the stolen materials was considered ‘fair use’, but the way they obtained the materials is nonetheless impermissible under copyright.

The ‘fair use’ doctrine doesn’t exist in Australian copyright law. Instead, the Copyright Act 1968 protects a narrower class of ‘fair dealing’ in copyrighted works, which is limited to a prescribed list of uses- including research and study, satire, and reporting the news. No Australian court has yet been asked to rule on whether a LLM being trained on copyrighted materials falls within a category of ‘fair dealing’, but it stands to reason that the narrower scope would create difficulty in arguing the case for an AI company. Even copyrighted texts used for AI ‘research’ will only be considered fair dealing if less than 10% of the work is used. Of course, owners of AI technology could avoid any possible legal implications under Australian copyright by compensating creators for the materials they use for LLM training.

Law policies such as the Australian Law Reform Commission have long advocated for the introduction of a ‘fair use’ provision into the Copyright Act. It is possible we will see this as a development of copyright law in Australia. For now, if a case like Kadrey or Bartz came before an Australian court, we would likely see a very different outcome.