Study on Development of Generative Artificial Intelligence from a Copyright Perspective

Generative artificial intelligence (GenAI) now underpins Europe’s creative and information industries, yet its dependence on vast, mostly copyrighted datasets exposes significant gaps in EU law. Commissioned by the EUIPO, this study applies a combined technical-legal-economic lens to test two propositions: whether current training practices remain inside the text-and-data-mining (TDM) exceptions, and how AI outputs can be kept both transparent and lawful.

The findings are stark. TDM is lawful only until a right-holder “expressly reserves” use, but no harmonised, machine-readable mechanism yet exists for doing so; creators rely on fragmented robots.txt instructions, bespoke metadata and platform clauses that web crawlers routinely overlook. At the output stage, watermarking and provenance initiatives for AI-generated content remain experimental and easily subverted. Despite this uncertainty, private actors are already striking high-value licensing deals for curated datasets, signalling an emerging market solution.

The report therefore recommends that the forthcoming EUIPO Copyright Knowledge Centre prioritise three actions: (1) build a central, machine-readable opt-out registry; (2) coordinate interoperable provenance standards for AI-generated material; and (3) publish model contracts and benchmarks that establish fair-value licensing norms. Together, these steps would align GenAI’s development with Europe’s commitment to both innovation and robust copyright protection.

The rapid growth of generative AI – which often relies on large volumes of copyrighted works for training data – has raised urgent legal and ethical questions in the realm of intellectual property. In response, the European Union Intellectual Property Office (EUIPO) commissioned a comprehensive study to clarify how generative AI systems interact with existing EU copyright law from technical, legal and economic perspectives. The study was conducted by a research team from the University of Turin Law School and the Nexa Center, and it forms part of the EUIPO’s broader effort to create a “copyright knowledge centre” in Europe.

The study was structured around three main components:

• A technical, legal and economic analysis to further understand the functionality of generative AI and the implications of its development.
• A detailed examination of copyright-related issues regarding the use of existing content in the development of generative AI services.
• A detailed examination of copyright issues regarding the generation of new content by generative AI systems.

The study found that no single solution or standard has yet emerged for rights holders to express reservations (opt-out) under the EU’s text-and-data mining exceptions, nor a universal transparency measure to identify and disclose AI-generated content. At the same time, the research noted that while the global generative AI landscape features an increasing number of legal disputes, there is also a trend of high-value licensing agreements being struck between rightsholders and AI developers (for example, deals between stock image libraries and major AI companies, or news publishers and AI model providers). The report suggests that public authorities may have a role to play in supporting technical solutions – such as managing databases of opt-out declarations – and in raising awareness of best practices to mitigate potentially infringing outputs. As a disruptive technology, generative AI is significantly changing the creative and IT industries and the ways in which rightsholders and AI developers operate. While it may take time for a new equilibrium to be established, the study underscored the importance of improving access to essential information about the origin and permitted uses of works, in order to ensure proper respect for copyright, fair benefit-sharing, and effective enforcement in the age of AI.