Transformative Use, The Black Box Problem, And The Unfinished Architecture Of AI Copyright Law
- IJLLR Journal
- Jul 8
- 2 min read
Akshit Mathur, Jindal Global Law School
ABSTRACT
The doctrine of transformative fair use has served as copyright law's principal mechanism for accommodating technological change. From the photocopier to the internet, courts have repeatedly held that repurposing copyrighted material for a fundamentally different function can excuse what would otherwise constitute infringement. Authors Guild, Inc. v. Google Inc. (2d Cir. 2015) represented the high-water mark of this approach: mass digitization of millions of books was deemed transformative because courts could inspect, measure, and verify the scope of Google's use through its snippet-display interface. A decade later, Bartz v. Anthropic PBC (N.D. Cal. 2025) asked whether the same logic could extend to large language models trained on millions of copyrighted works — and the answer revealed a doctrinal fault line that Authors Guild had papered over. This article argues that the two cases, read together, expose a structural vulnerability in the fair use framework: the doctrine can only function as intended when courts can empirically verify that a claimed transformation is genuine rather than merely asserted. In the age of opaque generative AI, that verification has become impossible. The article proceeds in three parts. Part I analyses the Authors Guild precedent and argues that its reasoning rested, implicitly, on the inspectability of Google's use. Part II examines how Bartz both extended and strained that precedent, with particular attention to the 'black box problem' — the inability of courts, authors, or even developers to trace how copyrighted material shapes model behaviour. Part III proposes a governance framework centred on mandatory dataset disclosure, third-party auditing, and collective licensing, arguing that these mechanisms are not merely desirable policy supplements but necessary prerequisites for the fair use doctrine to operate coherently in the context of generative AI.
