IP Law Daily, COPYRIGHT NEWS: Microsoft allegedly used thousands of books for AI training, authors sue, (Jun 26, 2025)

By Cathleen Calhoun, J.D.

A reasonable licensing fee could have been paid, or books in the public domain could have been used, but those options were not chosen, according to the complaint.

Copyright holders of a large array of works have filed a putative class action complaint against tech giant Microsoft for allegedly feeding hundreds of thousands of pirated books, without permission, into its large language model (LLM) to train its Megatron AI tool. The authors argue that they deserve compensation for their works and that the Copyright Act grants “a bundle of exclusive rights” to creators, including the right to reproduce the copyrighted works. They claim Microsoft infringed on their exclusive rights, and those of their proposed class, under 17 U.S.C. § 106 (Bird v. Microsoft Corp., No. 1:25-cv-05282 (S.D.N.Y. filed June 25, 2025)).

AI and LLMs. Recent generative AI systems, designed to recognize text that is inputted, and generate output text, are built on LLMs. The LLMs attempt to “understand” human language, and then mimic human language by generating output text on a predictive basis, such as predicting what word will follow another. Microsoft’s LLMs consist of a complex web of mathematical functions that break down input text into smaller pieces, or tokens. To train an LLM, large amounts of text must be put in for the LLM to ingest, and the more text the better. According to the complaint, the LLM “mines and feeds on the expression contained in the training corpus, adjusting algorithms such that it can mirror and mimic the ordering of words, style, syntax, and presentation of facts, concepts and themes.”

Alleged infringement. The complaint alleges that Microsoft trained its Megatron LLM on 25.7 billion tokens of a compilation of works called “Books3” and that “Books3” contained the works of the parties. An independent developer created Books3, according to the complaint, from a direct download of many books from a pirated website. Specifically, it was a compilation of 196,640 books, allegedly comprising all of Bibliotik, a “notoriously pirated collection” of “pirated books.” The authors allege that Microsoft deprived them of book sales and licensing revenue and usurped a licensing market for copyright owners. They state that the success of Megatron’s LLM was predicated on mass copyright infringement, and claim that Microsoft infringed on the exclusive rights, under 17 U.S.C. Sec. 106, of plaintiffs and members of their proposed class, by, among other things, reproducing the works owned by them, by downloading from pirated sources, and by training on reproduced copies of works.

Authors. Multiple examples of works used in Book3 are provided in the complaint. For example, Plaintiff Bird is the recipient of the 2006 Pulitzer Prize for Biography for American Prometheus: The Triumph and Tragedy of J. Robert Oppenheimer. A number of his books, including that one, were included in Books3. Some of the other works included in Books3 were also famous, including New York Times bestselling book Ghost Soldiers: The Epic Account of World War II’s Greatest Rescue Mission by plaintiff Sides. Fiction books were also part of Books3, including the romance novels written by one of the plaintiffs under the pen name Eloisa James. The copyright holders bring the suit in their name and in the name of the class of others who had their works allegedly downloaded by Microsoft.

Claim for relief. The plaintiffs demand a jury trial and request an award of actual damages. They also request statutory damages of up to $150,000 per infringed work, in the alternative to actual damages and profits, at their election before final judgment, among other requests.

The Case is No. 1:25-cv-05282.

Attorneys: Charlotte Lepic (Susman Godfrey LLP) for Kai Bird, Jonathan Alter, Mary Bly, Victor LaValle, Eugene Linden, Daniel Okrent, Hampton Sides, Jia Tolentino, Simon Winchester, and Eloisa James, Inc.

Companies: Microsoft Corp.; Eloisa James, Inc.

News: AINews Copyright TechnologyInternet NewYorkNews