Authors Say Microsoft Trained AI Using Their Pirated Works (1)

A group of authors allege Microsoft Corp. unlawfully used pirated versions of their copyrighted works to train its artificial intelligence large language model, according to a new federal lawsuit.

Microsoft used roughly 200,000 pirated books to train its Megatron-Turing Natural Language Generation model, they say in a complaint filed Tuesday in the US District Court for the Southern District of New York. The group alleges on behalf of itself and a proposed class that the technology company obtained the pirated works through a “notorious” collection titled “Books3.”

The success of Microsoft’s LLM “is predicated on mass copyright infringement,” the complaint says.

Microsoft didn’t immediately respond to a request for comment.

The group of authors— which includes former President Jimmy Carter biographer Jonathan Alter—alleges Microsoft intentionally chose to use pirated libraries, rather than licensing deals with publishers and creators, in an effort to gain “huge advantages” in its LLM. The alleged decision also keeps such illegitimate libraries in business and gives them a “seal of approval,” the complaint says.

Microsoft acknowledged that it used more than 800GB of open-source data called “The Pile” to train its LLM, the complaint says. The company allegedly utilized the dataset at a time when it contained “Books3.”

The pirated book collection was removed from the “most official version” of The Pile in 2023 due to copyright complaints, the group says.

The proposed class would include all owners of copyrighted works that registered with the US Copyright Office within five years of their work’s publication and that were downloaded or reproduced by Microsoft, the complaint says. Class members must also have registered their copyrighted works within at least three months of first publication, and also have either an international standard books number or an Amazon standard identification number.

The plaintiffs seek class certification; an injunction preventing Microsoft from infringing on their copyrights; actual damages; a statutory damages award of up to $150,000 per infringed work; pre- and post-judgment interest; and attorneys’ fees and costs, the complaint says.

Lieff Cabraser Heimann & Bernstein LLP; Susman Godfrey LLP; and Cowan DeBaets Abrahams & Sheppard LLP represent the authors.

The case is Alter v. Microsoft Corp., S.D.N.Y., No. 1:25-cv-05282, complaint 6/24/25.