The Impact of Generative Artificial Intelligence on Copyright Law and Intellectual Property Protection: A Comparative Doctrinal Analysis

Abstract

Generative artificial intelligence has placed copyright law under a kind of pressure it was not designed to absorb. Two questions sit at the centre of the problem. The first concerns inputs: whether copying millions of protected works to train a model is an infringing act or a lawful, even socially valuable, form of analysis. The second concerns outputs: whether text, images or music produced by a model can be owned at all, and by whom. This article examines both questions through a doctrinal and comparative study of the United States, the European Union, the United Kingdom and China, with briefer reference to Japan. It draws on statutes, regulatory guidance and the first wave of judicial decisions, including Thaler v. Perlmutter, Bartz v. Anthropic, Kadrey v. Meta, Getty Images v. Stability AI in the English High Court, GEMA v. OpenAI in Munich and Li v. Liu in the Beijing Internet Court. The analysis finds that courts now broadly agree that works lacking human creative control are unprotected in the United States, while the United Kingdom and China reach more permissive results by different routes. On training, the picture is less settled. American courts have so far treated training on lawfully acquired works as transformative, yet the Munich court held that a model which memorises lyrics itself contains a reproduction. Patent law is more uniform: every major office has refused to name an AI system as an inventor. The article argues that litigation alone cannot resolve these tensions and proposes a three-part reform built on training-data transparency, statutory remuneration for rights holders and a short neighbouring right for AI-generated output.