Publishers File New Copyright Lawsuit Against Google Over Gemini Training

Hachette Book Group, Cengage Learning, Elsevier, author Scott Turow, and S.C.R.I.B.E. filed a proposed class action against Google in federal court in New York. The complaint, filed on July 10, 2026 in case 26-cv-5870, alleges that Google copied books supplied for limited Google Books and related uses, scraped other works, removed copyright-management information, and used the material to develop Gemini models without permission. Hachette publicly announced the case on July 13, which explains why publication dates differ from the court filing date.

The plaintiffs ask the court for damages, an injunction, and destruction of copies they allege were made without authorization. They also seek to represent a broader class of authors and publishers. Proposed class status has not been granted, the allegations have not been adjudicated, and Google had not filed a court response when the reviewed reports were published.

The dispute is not simply about whether a model can learn patterns from text. It asks how the defendant allegedly obtained and copied the material, what rights accompanied each source, whether copyright-management information was removed, and whether any use is protected by fair use. Those questions are fact-specific and remain for the court.

The operational lesson is purpose limitation. A data pipeline should not treat a readable file, a search index, a purchase, or a partner feed as a universal training license. Every training record should preserve source URL, acquisition method, governing agreement, permitted purposes, copyright status, retention rule, transformation history, and deletion obligations.

Model teams also need a rights-aware data lineage graph. If a dataset is merged, filtered, tokenized, embedded, or distilled, the downstream artifacts should retain a link to the original rights record. That makes it possible to honor exclusions, respond to disputes, and prove which model versions were exposed to which source material.

LDS views this lawsuit as a chain-of-custody test for AI training data. The technical control is not a generic copyright flag; it is a purpose-bound authorization record enforced at ingestion and propagated through every derivative dataset. Organizations that cannot reconstruct that history will struggle to answer basic legal, security, and model-governance questions even before a court decides the fair-use issues.

The next material steps are Google's formal response, any motion addressing fair use or acquisition, decisions on class certification, and rulings on preservation or destruction of disputed copies. Until then, descriptions of infringement, intent, dataset contents, and damages must remain attributed allegations rather than established facts.