Abstract
Generative AI (GenAI) systems raise fundamental challenges for copyright law at both the input and output stages. On the input side, legal uncertainty surrounds the large-scale scraping of copyrighted data for model training, with divergent rules across jurisdictions and limited transparency on how data is sourced. On the output side, courts struggle to determine when AI-assisted creations are sufficiently human to merit protection, leading to inconsistent or unclear legal outcomes. This paper outlines the "AI copyright conundrum" and examines its impact on the incentives to create, the accessibility of high-quality datasets, and the sustainability of cultural production. We discuss policy options and open questions for research.