TRANSPARENCY OBLIGATIONS AND COPYRIGHT COMPLIANCE OF GENERAL-PURPOSE AI MODELS UNDER EU LAW

The fast growth of artificial intelligence technologies, especially general-purpose AI (GPAI) models, has seriously challenged traditional methods of copyright protection in the European Union.GPAI models are trained on large datasets that often include works protected by copyright, databases, and computer programs.This raises complicated legal issues about lawful use, authors' rights, and the responsibility of AI developers.Current copyright rules, like the text and data mining (TDM) exceptions introduced by Directive (EU) 2019/790 on Copyright in the Digital Single Market, were not meant to handle the scale, lack of clarity, and cross-border nature of modern AI training practices.The EU Artificial Intelligence Act marks a significant regulatory change aimed at addressing these issues.Specifically, Article 53 of the AI Act establishes binding rules for transparency and copyright compliance for providers of general-purpose AI models.These rules require GPAI providers to create internal policies for copyright compliance, respect rightsholders' opt-out requests under Article 4(3) of the DSM Directive, and publish clear summaries of their training datasets.This approach aims to improve accountability and help enforce intellectual property rights without requiring full disclosure of proprietary datasets.The article looks at how the AI Act interacts with the EU copyright framework, focusing on how effective the transparency obligations are as a regulatory tool.It investigates whether the requirement to summarize training data truly strengthens the position of rightsholders and compensates for the weaknesses of the opt-out-based TDM system.It pays particular attention to the conflict between transparency requirements and the protection of trade secrets, as well as the effects of applying EU