Kadrey v. Meta Platforms, Inc.
3:23-cv-03417
| N.D. Cal. | Jun 25, 2025Background
- Thirteen authors sued Meta for using their copyrighted books, obtained from shadow libraries, to train its Llama large language models (LLMs) without permission or licensing.
- Meta initially sought to license books for AI training but abandoned negotiations after facing logistical issues and instead procured books via shadow library downloads.
- The plaintiffs primarily asserted that Meta's conduct could not be considered fair use and was therefore infringing, focusing on market harm from reproduction of text snippets and loss of licensing opportunities.
- Meta did not dispute copying but argued its use was highly transformative and thus eligible for the fair use defense.
- Both sides moved for partial summary judgment on fair use; the ruling concerns only the individual claims of these thirteen plaintiffs, not potential class members.
- The court found that, while unauthorized AI training with copyrighted works often poses serious market and copyright concerns, the plaintiffs in this case failed to present sufficient evidence of market harm from Meta's use.
Issues
| Issue | Plaintiff's Argument | Defendant's Argument | Held |
|---|---|---|---|
| Fair use - Transformative Use | LLMs have same purpose as original works; no critical transformation | Use is highly transformative (AI training vs. reading); LLM purpose is distinct | Use is highly transformative, favoring Meta on factor one |
| Fair use - Nature of Copyrighted Work | Books are highly creative, deserve strong copyright protection | Use targets non-expressive, statistical patterns, akin to functional elements | Works are highly creative, factor favors plaintiffs but is minor |
| Fair use - Amount/Substantiality Used | Entire books were used, exceeds fair use | Copying entire books is necessary for effective AI training | Amount copied reasonable in relation to transformative purpose, favors Meta |
| Fair use - Market Harm | LLM regurgitates text; loss of licensing opportunities for AI training; dilution | No meaningful regurgitation; right to training-data licensing doesn't exist; dilution speculation | Plaintiffs failed to present concrete evidence of market harm; theory of market dilution plausible but not established in record |
| Acquisition from Shadow Libraries | Downloading from shadow libraries (piracy) precludes fair use | Source of books is not legally dispositive; transform. purpose remains | Bad faith and piracy allegations not dispositive; transformative use and overall harm are more important |
Key Cases Cited
- Twentieth Century Music Corp. v. Aiken, 422 U.S. 151 (copyright's purpose is public access to the arts)
- Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (fair use is highly fact-specific; transformative use analysis)
- Google LLC v. Oracle America, Inc., 593 U.S. 1 (adaptation of fair use to technological change; transformative use)
- Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (market harm is key factor in fair use)
- Harper & Row Publishers, Inc. v. Nation Enterprises, 471 U.S. 539 (market effect as crucial fair use factor)
- Authors Guild v. Google, Inc., 804 F.3d 202 (transformative use and market harm in digital context)
- Sega Enterprises Ltd. v. Accolade, Inc., 977 F.2d 1510 (intermediate copying, functional element analysis)
- Mattel, Inc. v. MGA Entertainment, Inc., 616 F.3d 904 (protection of expression over style)
- Disney Enterprises, Inc. v. VidAngel, Inc., 869 F.3d 848 (limits on intermediate/functional analysis for fair use)
- Kelly v. Arriba Soft Corp., 336 F.3d 811 (amount copied reasonable if related to transformative purpose)