Nvidia rejected claims that its AI was trained on illegal copies of books

New Nvidia Position in the Pirate Book Use Case

In a federal court in California, NVIDIA responded to a lawsuit alleging that its AI models were trained using content from “shadow” libraries, including Anna’s Archive and Books3. In a statement dated January 29, NVIDIA declined to acknowledge any evidence of copyright infringement.

What the Company Claims
1. Lack of Specific Facts

– Plaintiffs (Nazemian vs. Nvidia) failed to show which of their works were uploaded or used in training NVIDIA models.

– Even basic elements required for a copyright claim are missing: no information on copying, where it occurred, or which models contained protected material.

2. “Anna’s Archive” Section

– The lawsuit describes internal discussions among NVIDIA employees about possible access to a pirated site. However, the company stresses that this does not prove actual downloading or use of content.

– According to Nvidia, evaluating potential data sources is not equivalent to copying protected works.

3. Critique of Plaintiffs’ Grounds

– Plaintiffs rely on “talks and beliefs” rather than concrete evidence. Nvidia views this as an attempt to use discovery as a basis for the claim, which copyright courts typically deem inadmissible.

4. Narrowing the Dispute Scope

– The complaint adds information about additional datasets and models (e.g., Megatron 345M). Nvidia argues that plaintiffs improperly bundle multiple products in one case without explaining how each model was trained on their works.

5. Theory of Secondary Liability

– The lawsuit references the NeMo Megatron framework and its support for large public datasets (The Pile). Nvidia notes that the statement does not specify direct infringement by third parties—an essential element for asserting indirect liability.

How the Motion Looks
- Overall Position – Claims contain only speculative assumptions without concrete facts.

- Evidence Requirements – The court must see proof of copying and its circumstances.

- Summary – NVIDIA requests dismissal because the claim fails to meet copyright law requirements.

Conclusion
Nvidia’s statement clearly rejects acknowledging any evidence of copyright infringement by the company. The case remains pending in the Northern District of California, where the judge will decide whether the presented allegations suffice for a valid intellectual property claim.