Letting sleeping wasps lie: general-purpose AI models and copyright protection under the European Union AI Act

Abstract This article addresses two principal research objectives: first, to examine how and to what extent the provisions of the EU AI Act (EUAIA) dedicated to general-purpose artificial intelligence (AI) models (GPAIm) govern the intersection of copyright and AI, through the scrutiny of major legal challenges raised by AI manipulations of copyrighted material, and, second, to distil whether and how the EUAIA’s provisions on prohibited AI practices can be applicable to AI-based copyright infringement. The author posits that the provision of Article 5(1)(a) EUAIA, which does not explicitly refer to copyright, can be interpreted in a sense that it allows us to qualify the use of copyrighted material for creating manipulated content as a ‘purposefully manipulative or deceptive technique’, but only once the other criteria of Article 5(1)(a)—not related to copyright law—are fulfilled. A specific, customized methodological approach with three main characteristics is adopted to successfully address the objectives outlined above. First, the legal content analysis of the EU legislation is teleological, dynamic and holistic. Second, the literature review is combined with references to relevant jurisprudence and insights into legal reasoning of policy makers and practitioners. Third, the methodological approach is, whenever necessary, based on convergence of legal studies with interdisciplinary explorations of political science, psychology, economics and technologies. The article argues that, by refraining from a more comprehensive regulation of the intersection between GPAIm and copyright law, the EU legislators have inadvertently created a fertile environment for a multitude of interpretative and enforcement challenges of the Union’s regulatory framework in the Member States.