In re Salesforce LLM Copyright Infringement Litigation

On 15th October 2025, two authors, E. Molly Tanzer and Jennifer Gilmore, filed proceedings against Salesforce, Inc. in the United States District Court for the Northern District of California.

According to the plaintiffs, Salesforce’s models were trained in part on Books3, a dataset sourced from the now-defunct pirate site Bibliotik. Books3 has been used as part of broader training datasets such as The Pile and RedPajama, which are known to contain thousands of copyrighted literary works. The Plaintiffs allege that their books were among the materials used without consent.

The Plaintiffs allege that Salesforce initially listed RedPajama-Books among the datasets used to train some of its models but later updated its documentation to describe the training data more generally as “public and legally available.” The Plaintiffs assert that this change raises concerns about transparency and the lawful sourcing of training data. The models at issue include XGen-7B, xGen-Sales, and xGen-Small, which are used within Salesforce’s Agentforce AI platform.

The Plaintiffs seek statutory damages and class certification for all U.S. authors whose copyrighted works may have been included in the model training data without permission. The case joins a growing number of lawsuits addressing the use of copyrighted materials in generative AI training arising from the use of materials from sources such as The Pile.

This summary is provided for educational and research purposes only and does not constitute legal advice. Allegations described are those made in the filings. Please consult the underlying source documents.