OpenAI accused of keeping pirated book library in unsealed filing

OpenAI built and kept a centralised library of pirated books for uses that went beyond training data, book plaintiffs allege in newly unsealed court papers. The papers include a motion for summary judgment, meaning a request for a judge to decide claims without a full trial. The filing was described on 21 September 2026 in trade coverage of the case Publishers Marketplace.

The core claim is about source material and storage. Plaintiffs allege the company drew on pirated copies and then organised them for use and sharing outside model training. That distinction matters in publishing. Training is one alleged use. A standing library is another.

Internal messages sit at the centre of the account. An unsealed filing details staff discussion of taking books from Library Genesis, a website known for hosting pirated copies, for OpenAI work The Wall Street Journal. The messages use blunt, informal language. They are presented as evidence of how staff viewed the practice.

The motion also points to internal documents about risk to books. In those documents, employees are said to have believed that large language models, the software behind chatbots such as ChatGPT, would create substitutes for books and threaten creative output. For authors, that is familiar ground. Substitution is the commercial fear behind many copyright claims.

Procedure is now part of the fight. In Authors Guild et al v. OpenAI Inc. et al, No. 1:2023cv08292, the plaintiffs seek a ruling on whether OpenAI must hand over discovery, meaning documents and communications exchanged before trial, about its deletion of datasets known as Books1 and Books2. Deletion is important because it can affect what evidence remains. The court must decide what has to be produced.

The dispute sits inside wider litigation in the United States. In Re: OpenAI, Inc. Copyright Infringement Litigation, No. 1:2025md03143, a copyright case filed on 11 April 2025 in the U.S. District Court for the Southern District of New York, groups related claims together. Two cases listed in that district show the range. Chabon et al v. OpenAI, Inc. et al, No. 1:2025cv03291, was filed on 21 April 2025. Millette v. OpenAI, Inc. et al, No. 1:2025cv03297, names David Millette and Ruslana Petryazhna as plaintiffs and OpenAI Inc., OpenAI OpCo LLC and Microsoft Corporation as defendants.

What makes this stand out for publishing is the chain it describes. A manuscript becomes a book, a book becomes a file, a file enters a library, and that library feeds a system that can write like a book. Each step has different owners and different terms of trade. The court will now test which steps needed permission.