COPYRIGHT PROTECTION IN THE CONTEXT OF THE PROVISION OF AI MODELS FOR GENERAL USE

This paper examines how the new European AI Act (Regulation (EU) 2024/1689) governs the use of copyrighted content in the training and deployment of general-purpose AI models, focusing on the pivotal Article 53. Article 53 is a key provision that imposes explicit obligations on AI model providers to ensure transparency, rigorous documentation, and adherence to copyright law. The analysis highlights the legal and ethical implications of using copyrighted works to train AI systems, noting the tension between the need for vast datasets and the rights of authors. It explores how providers are required to maintain detailed technical documentation and publicly disclose summaries of training data, implement policies to comply with copyright (including honoring rights-holders’ opt-outs), and guarantee compliance with EU copyright and related rights. These responsibilities aim to increase accountability and enable oversight while fostering trust in AI outputs. At the same time, the paper discusses how the regulation seeks to balance innovation and copyright protection, guided by principles of proportionality and fairness: general-purpose AI development is permitted but constrained to respect authors’ economic and moral rights. The broader EU legal context – including existing copyright exceptions (such as text and data mining allowances), moral rights of creators, and the three-step test – is considered to understand the boundaries of lawful AI training. Finally, the abstract addresses enforcement challenges, such as ensuring transparency without compromising trade secrets, difficulty in tracking protected content in large datasets, and cross-border compliance. The study concludes that Article 53 represents a significant step toward aligning AI innovation with European copyright norms, striking a delicate balance between fostering technological progress and safeguarding intellectual property rights.