Tanzer v. Adobe — AI Copyright Claim Fails Without Facts Tying Books to Training Data

Background

Author E. Molly Tanzer alleged that Adobe unlawfully used her copyrighted books while developing generative-artificial-intelligence products. The challenged portion of her claim concerned Nemotron, a family of language models associated with NVIDIA. Tanzer alleged generally that AI developers commonly obtain books from unauthorized “shadow libraries,” and that her works were available through Anna’s Archive.

Adobe moved to dismiss the Nemotron portion of the case. The immediate question was not whether training an AI model on copyrighted books can infringe copyright, but whether this complaint pleaded enough facts to support a reasonable inference that Tanzer’s own works were used in the training process connected to Adobe.

The Court’s Holding

The court dismissed the Nemotron theory with leave to amend. It held that Tanzer had not plausibly alleged either Article III standing or direct infringement because the complaint did not connect Nemotron’s training data to Anna’s Archive and, in turn, to her books.

General allegations that piracy is common in AI development did not fill that gap. The court reasoned that repeating a conclusion—that NVIDIA or Adobe used Anna’s Archive—does not make the conclusion factual or plausible. Even drawing inferences for Tanzer, the allegations did not show that Nemotron was trained on the repository containing her works. The court allowed an amended complaint, giving Tanzer an opportunity to add concrete facts by October 2, 2026.

Key Takeaways

• AI-copyright plaintiffs must plausibly trace their own works into a defendant’s training pipeline; industry-wide allegations are not enough.
• Availability of a book in a shadow library does not itself establish that a particular model used that library.
• The ruling addresses pleading and standing, not the ultimate legality of training generative AI on copyrighted material.

Why It Matters

The decision illustrates the increasingly important dataset-attribution problem in AI copyright litigation. Authors may suspect that their works entered model training through large illicit collections, but a complaint still needs facts linking the particular model, dataset, repository, and copyrighted work. For AI developers, the ruling reinforces the litigation value of documented data provenance. For creators, it shows why discovery access and reliable dataset evidence can determine whether a claim reaches the merits.