Public AI and Copyright
Abstract
Public Artificial Intelligence (AI) initiatives seek to combine public accountability, public interest objectives and open or equitable access to AI infrastructures and models. Their development nevertheless depends on legal frameworks that do not recognise Public AI as a distinct category. This article examines how European Union copyright law and the AI Act enable and constrain Public AI development and dissemination. It combines doctrinal and policy analysis with a documentary survey of European initiatives and case studies of GPT-NL and the Norwegian National Library's licensing arrangements with Kopinor. The analysis identifies meaningful but unevenly available legal space. Article 3 of the Copyright in the Digital Single Market Directive provides a mandatory, unremunerated text and data mining exception for qualifying institutions conducting scientific research. But uncertainty persists concerning lawful access, collaborative development and the dissemination or deployment of resulting models. The AI Act adds copyright compliance and training data transparency obligations, while its research exclusions and open-source exemptions provide limited flexibility. The case studies reveal contrasting approaches to dataset access, licensing, remuneration and downstream use. They also expose a tension between negotiated licensing and the practical effectiveness of the research exception. The article argues for targeted clarification of the existing framework rather than a separate Public AI regime. Licensing should facilitate access and authorise uses outside Article 3 without becoming a condition for activities already permitted by it. Clearer coordination with the AI Act is also needed to support collaborative research and model dissemination.
Links
Artificial intelligence, Copyright