IP Law Daily, COPYRIGHT NEWS: Authors sue Anthropic for infringement in ‘training’ of AI model, (Aug 20, 2024)

By Karen Servidea, J.D.

The authors allege that AI company Anthropic improperly copied their works in datasets used to train the company’s LLM.

Three book authors have filed a putative class action copyright infringement lawsuit in the Northern District of California against Anthropic PBC, the company behind the “Claude” series of AI tools. The complaint alleges that the AI company downloaded known pirated copies of the authors’ works, made copies of them, and used them to train the company’s “large language model” (LLM), all without paying for even an initial copy of the works. The named plaintiffs are Andrea Bartz, author of The Lost Night: A Novel; Charles Graeber, author of The Good Nurse: A True Story of Medicine, Madness, and Murder, and The Breakthrough: Immunotherapy and the Race to Cure Cancer; and Kirk Wallace Johnson, author of To Be a Friend Is Fatal: The Fight to Save the Iraqis America Left Behind (Bartz v. Anthropic PBC, No. 3:24-cv-05417 (N.D. Cal. filed Aug. 19, 2024)).

According to the complaint, Anthropic was founded in January 2021 by seven former OpenAI employees and released the first iteration of its flagship model Claude in March 2023. Again, according to the complaint, Anthropic released subsequent iterations thereafter, with “Claude 3.5 Sonnet” being released in June 2024.

The complaint explains that LLMs attempt to “understand” human language by processing input text, and are designed to generate output text, mimicking human language usage, on a predictive basis. The complaint describes Claude as a complex web of mathematical functions comprised of a series of algorithms that break down input text into smaller pieces or “tokens” and then translate those tokens into number sequences, which help place the tokens on a map by identifying other closely associated tokens. “As the model trains and digests more expression, the algorithms depicting the relationship between various tokens changes with it.” According to the complaint, the model is trained on a massive body of text and without training there would be no LLM. The complaint posits that books are especially valuable training material.

In that regard, the authors allege that to train Claude, Anthropic copied a massive trove of pirated books, including books belonging to the plaintiff authors. In particular, according to the complaint, a non-party created “Books3,” a direct download of books from a pirated website with roughly 197,000 pirated books. Another non-party then allegedly included Books3 in a larger dataset called “The Pile.” The authors allege that the defendant used The Pile dataset as a source for its training corpus for Claude.

The complaint asserts that Anthropic thus copied and exploited copyrighted books in knowing violation of copyright laws, specifically their exclusive rights to reproduction and distribution under Section 106 of the Copyright Act, 17 U.S.C. § 106. The complaint posits that Anthropic deprived the authors of books sales and licensing revenues. It further argues that Anthropic profited from the unlicensed exploitation of the copyrighted material at the expense of the authors. In that connection, the complaint asserts that Claude and other LLMs threaten the livelihood of the authors on whose works they were trained and that Claude in particular has been used to generate cheap book content.

The authors seek an unspecified amount of statutory or compensatory damages at the plaintiffs’ election, restitution, disgorgement, attorney fees and costs, and a permanent injunction against the alleged infringement.

The case is another among several copyright lawsuits involving generative AI. For example, a group of authors filed suit against OpenAI in the Southern District of New York in September 2023. OpenAI was also named as a defendant, along with Microsoft, in a complaint filed in the Southern District of New York in November 2023. Anthropic itself was named as the defendant in a complaint alleging infringement filed by music publishers in the Middle District of Tennessee in October 2023.

The Case is No. 3:24-cv-05417.

Judge: Alsup, W.

Attorneys: Justin A. Nelson (Susman Godfrey L.L.P.) for Andrea Bartz.

Companies: Anthropic PBC

News: AINews Copyright TechnologyInternet CaliforniaNews GCNNews