The "Fair Use" and "Fair Dealing" Dilemma in Large Language Model Pre-Training: A Comparative Analysis of US, EU, and UK Copyright Frameworks

Abstract

The deployment of large language models (LLMs) such as GPT-4, Claude, and Gemini has precipitated a global legal crisis in copyright law, centred on a deceptively simple question: does the wholesale ingestion of copyrighted text to train neural networks constitute permissible use or wholesale infringement? This paper provides a systematic comparative analysis of the three most consequential legal frameworks governing this question — the United States fair use doctrine under 17 U.S.C. § 107, the European Union's text and data mining (TDM) exceptions under Articles 3 and 4 of the Digital Single Market (DSM) Directive (Directive 2019/790/EU), and the United Kingdom's Copyright, Designs and Patents Act 1988 (CDPA), as amended and currently under consultation. Drawing on landmark judicial decisions — including Bartz v. Anthropic (N.D. Cal. 2025), Kadrey v. Meta (N.D. Cal. 2025), The New York Times Co. v. Microsoft Corp. (S.D.N.Y. 2023), Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc. (D. Del. 2025), and GEMA v. OpenAI (LG München I, 2025) — as well as legislative instruments including the EU AI Act (Regulation 2024/1689) and the UK Government's March 2026 Report on Copyright and Artificial Intelligence, this study maps the points of structural convergence and persistent divergence among the three regimes. The analysis reveals that while all three jurisdictions nominally balance innovation interests against creators' rights, their operational mechanisms are fundamentally different in design: the US adopts a post-hoc, litigation-driven four-factor balancing test; the EU employs an ex ante, legislatively structured opt-out framework; and the UK remains in a state of policy flux, unable to definitively settle between the two models. This article proposes that the emerging international standard should synthesise the EU's structural clarity with US jurisprudential flexibility to produce a regime that is both commercially workable and normatively defensible.