Post-Warhol Transformative Use
Recalibrating 17 U.S.C. § 107 in the Age of Synthetic Content
The Supreme Court's 2023 ruling in Andy Warhol Foundation v. Goldsmith marked a historic structural pivot in American fair use jurisprudence. This paper traces the thirty-year evolution of the Transformative Use doctrine from Judge Leval's 1990 essay through Campbell, Google v. Oracle, and Warhol, demonstrating how the new test alters the viability of fair use defenses for generative AI foundation models.
I. The Rise of the Transformative Hegemony
For three decades following Campbell v. Acuff-Rose (1994), lower federal courts treated 'transformative use' as a virtually dispositive trump card under Section 107. If a defendant demonstrated that their work added new aesthetic meaning or social commentary, Factor 1 overwhelmingly weighed in their favor, carrying the overall fair use determination even in heavily commercial settings.
This expansive interpretation culminated in the Second Circuit's ruling in Cariou v. Prince (2013), which held that artist Richard Prince's alterations to Patrick Cariou's photographs of Rastafarians were fair use as a matter of law simply because Prince created a different 'mood' and aesthetic.
II. Sotomayor's Course Correction: The Specific Transaction Rule
Justice Sotomayor's majority opinion in Warhol v. Goldsmith decisively rejected the Cariou doctrine. The Court established that transformativeness is not a generalized metaphysical inquiry into an artwork's meaning, but an objective analysis of the specific commercial transaction at issue.
Because both Lynn Goldsmith's original photograph and Andy Warhol's 'Orange Prince' silkscreen were licensed for the exact same commercial purpose (illustrating a magazine story about Prince), the Foundation's use lacked independent transformative justification. To hold otherwise, Sotomayor warned, would destroy the copyright owner's statutory exclusive right to prepare derivative works under 17 U.S.C. § 106(2).
III. The Crisis for Foundation AI Models
Warhol's narrowing of Factor 1 deals a severe blow to AI developers arguing that foundation model training is inherently fair use under Authors Guild v. Google. While Google Books indexed text to provide search metadata without competing in the book reading market, generative AI models are trained on expressive text and art to produce directly competing expressive text and art.
When an AI system is deployed commercially to generate synthetic journalism, illustrations, or music that directly substitute for the human creations ingested during pre-training, Warhol instructs courts that Factor 1 and Factor 4 merge to defeat fair use.
Precedential Parallels Engine
Tracing modern AI copyright disputes back to their historical technological counterparts: cameras, VCRs, search crawlers, and software APIs.
Analytic Note: Parallels indicate similarity in underlying legal and technological mechanisms, not factual equivalence. These historical lineages illuminate how courts historically adapted copyright doctrines to technological shocks.
When photography was challenged as purely mechanical recording incapable of human authorship, SCOTUS held that mental conception, staging, lighting, and posing supply the requisite creative spark.
“Photographs are useful, new, harmonious, characteristic, and graceful pictures, so far as they are representatives of original intellectual conceptions of the author.”— 111 U.S. 53 (1884)
Judges are now evaluating whether prompt engineering, ControlNet constraints, and seed manipulation constitute the modern equivalent of Sarony's manual lighting and drapery arrangement.