NVIDIA faces lawsuit over AI training with copyrighted books
Multiple authors brought a class action copyright lawsuit against NVIDIA, alleging the company trained its NeMo Megatron AI framework using 197,000 unauthorized e-books from the Books3 dataset. A federal court judge rejected NVIDIA's motion to dismiss, finding that specific data processing scripts in the framework had the sole purpose of accelerating copyright infringement and were not protected as general-purpose development tools.
What happened
Multiple authors filed a class action copyright lawsuit against NVIDIA, alleging the company used 197,000 unauthorized e-books from the Books3 dataset to train its NeMo Megatron large language model framework. The Books3 dataset was integrated into The Pile, an 800GB open-source text dataset, which was then used for model training.
NVIDIA sought dismissal citing internet service provider safe harbor protections under U.S. copyright law and claiming NeMo Megatron is a general-purpose AI training tool with substantial non-infringing uses. A federal judge in the Northern District of California rejected all of NVIDIA's dismissal arguments, finding that specific data processing scripts within the framework had the sole purpose of accelerating infringement and were not equivalent to general-purpose development tools.
NVIDIA was also alleged to have distributed scripts and tools to customers including Writer, Persimmon AI Labs, and Amazon that automatically downloaded The Pile, providing infrastructure for third-party infringement.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.