Lawsuit tracker
Nazemian v. NVIDIA
Nazemian v. NVIDIA Corp., No. 24-cv-01454 (N.D. Cal.)
| Plaintiffs | Authors Abdi Nazemian, Brian Keene and Stewart O’Nan, later joined by Andre Dubus III and Susan Orlean, on behalf of a proposed class |
|---|---|
| Defendants | NVIDIA Corporation |
| Court | U.S. District Court for the Northern District of California (Judge Jon S. Tigar) |
| Filed | 2024-03-08 |
| Status | Active |
| Content type | Books — nearly 200,000 titles in the Books3 collection within The Pile, sourced from the shadow library Bibliotik |
| Last updated | 2026-07-21 |
The claims
Direct and contributory copyright infringement over training NeMo Megatron models on The Pile, including Books3; a vicarious infringement claim was dismissed with leave to amend
What has happened
Three novelists sued NVIDIA in March 2024, alleging its NeMo Megatron models were trained on The Pile, which includes the Books3 collection of nearly 200,000 books copied from the shadow library Bibliotik. A related author suit was consolidated into the case, adding Andre Dubus III and Susan Orlean as plaintiffs. On May 5, 2026, Judge Jon Tigar denied NVIDIA’s motion to dismiss in large part, holding the authors plausibly alleged direct infringement and that scripts in NVIDIA’s NeMo framework — which plaintiffs say served to speed up the downloading of pirated books — could support contributory liability. The court rejected NVIDIA’s attempt to compare itself to a passive internet service provider. A vicarious infringement claim was dismissed with leave to amend. In spring 2026, the court also pressed NVIDIA to produce discovery about the shadow-library datasets used in training.
Key developments
- 2024-03-08 — Nazemian, Keene and O’Nan file a proposed class action over training NeMo Megatron on Books3; a related suit by Dubus and Orlean is later consolidated into the case.
- 2026-05-05 — Judge Tigar denies NVIDIA’s motion to dismiss in large part: direct and contributory infringement claims proceed, while vicarious infringement is dismissed with leave to amend.
Why it matters for training data
NVIDIA is best known as a supplier, and this case extends training-data liability into the toolchain. The court let a contributory claim proceed partly on allegations that NeMo framework scripts facilitated infringing downloads — a theory that reaches vendors of data-preparation tooling, not just the party running the training job. It is also another major case built on Books3, confirming that one pirated corpus can seed claims across many defendants. Suppliers shipping dataset tooling should assume plaintiffs will read their code.
Sources
- Courthouse News: NVIDIA cannot shake authors’ claims it trained AI on pirated books
- Loeb & Loeb: Nazemian v. NVIDIA Corporation case summary
- Tom’s Hardware: NVIDIA’s ISP defense fails as court declines to dismiss book-piracy claims
- CourtListener docket: Nazemian v. NVIDIA Corporation
Deeper analysis
Want data that clears this in diligence?
Whether you're building a model or sitting on an archive, the first conversation is short and specific.
Send a brief