Court narrows the consolidated Google AI training case: claims over Gemini, Imagen, PaLM, LaMDA, Bard and GLaM proceed

What happened

On 11 September 2025, Judge Eumi K. Lee of the U.S. District Court for the Northern District of California issued an order in In re Google Generative AI Copyright Litigation (Master File Case No. 5:23-cv-03440-EKL), the consolidated class action in which authors and visual artists allege Google trained its generative AI models on their copyrighted works without permission. The case combines Leovy v. Google (filed 2023, by writers) and Zhang v. Google (filed 2024, by photographers, cartoonists and illustrators including Jingna Zhang and Sarah Andersen).

The order granted Google's motion to dismiss in part, without leave to amend. Direct infringement claims were dismissed with prejudice as to ten models — Codey, Chirp, Veo, MedLM, LearnLM, SecLM, Gemma, CodeGemma, RecurrentGemma and PaliGemma — because the court found the complaint did not connect the plaintiffs' works to those models. The vicarious infringement claims against Alphabet Inc. were also dismissed with prejudice. The direct infringement claims proceed for six products: PaLM, GLaM, LaMDA, Bard, Gemini and Imagen.

The complaint alleges the text models were trained on datasets including Google's C4, which plaintiffs say contains material collected from pirated book sites, and that Imagen was trained on the LAION-400M dataset and likely LAION-5B. The plaintiffs allege copying in the training process itself; notably, they do not allege that the models produce infringing output.

Why it may matter to you

This is the main consolidated U.S. case about Google's training data for its language and image models. After this order, the case turns on specific datasets — C4/Infiniset for text and LAION datasets for images — so whether a creator's work is in those datasets is central to who the case can cover.

What is not yet known

The order decided only which claims were plausibly pleaded; it did not decide whether training on the works was infringement or fair use. The plaintiffs filed a motion for class certification in October 2025, which had not been decided in the documents reviewed here. No trial date or merits ruling is recorded in this source.