A Critical Analysis on Generative AI Model Training with Reference to Section 107 of the US Copyright Act, 1976
Copyright law was built for an analog world of tangible replication and human authorship. Generative Artificial Intelligence (GenAI) models destroy such architecture due to these models being trained by ingesting mass amounts of data, consisting mostly of copyrighted material, and are used without obtaining any sort of permission or authorisation for the same. This paper seeks to act as a re-evaluation of the same with reference to Section 107 of the US Copyright Act of 1976, dealing with the doctrine of Fair Use, whose evolution can be traced from the case of The Authors Guild, Inc. v. Google, Inc. (2015), through Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith (2023), to the recent 2025 judgments pertaining to the cases of Kadrey v. Meta Platforms, Inc.; Bartz v. Anthropic PBC; and Thomson Reuters Enterprise Centre GmBH v. ROSS Intelligence Inc., and along being read with the case of Thaler v. Perlmutter, regarding human authorship. Adopting a doctrinal methodology for research, this paper argues that courts protect the training process for these models, but begin to collapse once the result is provided, which consists of ingested artificial material, settling on the market substitution standard. Further argument provided in the paper is about the four-factor test provided by the doctrine cannot differentiate between the legal acquisition of such data and computational training, thereby evaluating collective licensing as a structural response.
Authors
- Mohamed Arfan Salih F
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24413931
- Primary Topic
- Law, AI, and Intellectual Property
- Type
- article
- Field-Weighted Citation Impact
- 0.00