Major publishers have sued Google in a New York federal court for allegedly training its Gemini AI on copyrighted books without permission.
The lawsuit highlights a growing tension between the generative AI industry and content creators over the legality of using intellectual property for machine learning. If the court finds that training AI on copyrighted works without a license constitutes infringement, it could fundamentally change how AI models are developed and monetized.
Three publishers — Hachette, Cengage, and Elsevier [1] — filed the suit in the U.S. District Court for the Southern District of New York [2]. The plaintiffs allege that Google secretly utilized millions of copyrighted books [3] to refine the capabilities of Gemini.
According to the filings, the companies said that this unauthorized use of their catalogs constitutes a direct violation of copyright law [4]. The publishers seek damages and a resolution to the unauthorized ingestion of their protected texts into Google's training datasets [1].
Google has not yet provided a detailed public defense against these specific allegations in the court filings. However, the case follows a broader trend of litigation where authors and publishers challenge the "fair use" doctrine as it applies to large language models [3].
The dispute centers on whether the process of "training" — where an AI analyzes patterns in text to predict the next word — creates a derivative work or simply learns a concept. The publishers said that the scale of the data used is too vast to be considered fair use [4].
“Major publishers have sued Google in a New York federal court for allegedly training its Gemini AI on copyrighted books without permission.”
This litigation represents a critical test for the 'fair use' defense in the age of generative AI. If the court rules in favor of the publishers, AI companies may be forced to negotiate expensive licensing deals for training data, potentially slowing the development of LLMs or increasing the cost of AI services. Conversely, a win for Google would solidify the practice of using public-facing internet data for training without compensation.
