In recent years, generative artificial intelligence has raised new legal questions at the intersection of technological development and copyright law. Systems capable of generating text, images, music or software code rely on vast training datasets and complex machine-learning models. At the same time, these systems raise fundamental questions about how traditional copyright concepts apply in a digital environment in which content is often generated by algorithms.
At the forefront is the question of whether, and to what extent, generative artificial intelligence interferes with the rights of copyright holders and what role human creative contribution plays in content produced with the assistance of such technologies.
Fundamental concepts of copyright law in the context of AI
Traditionally, copyright law protects a copyright work as an original expression reflecting the author’s own intellectual creation. The key criterion is therefore human creative contribution. In the context of content created with the assistance of artificial intelligence, the question arises whether human creative influence still predominates in the final result or whether the decisive aesthetic and creative choices are made by the system.
Among the core rights of copyright holders are the right of reproduction and the right of making a work available to the public, both of which are particularly significant in the digital environment. Generative models typically rely on large volumes of data for their operation, which may include copyright-protected works.
In this context, the concept of text and data mining (TDM) is also important. TDM enables the automated analysis of large volumes of text and data. Under European law, reproductions for the purpose of TDM are permitted in certain circumstances, provided that lawful access to the content exists and that the rights holder has not expressly reserved their rights (the so-called opt-out mechanism).
From training data to generated output
In generative AI systems, legal issues arise at several stages. First, at the stage of collecting and using training data, where the key question is whether protected works were lawfully used to train the model. Secondly, at the level of the model itself, where the question arises whether storing statistical patterns could amount to reproduction of a protected work if its expression can be reproduced or retrieved.
A further aspect concerns the user prompt, by which a user initiates the generation of content. Legal questions arise as to whom the relevant conduct should be attributed—whether to the user or the system provider—and whether a particular prompt may target the retrieval of protected expression.
Finally, the generated output is also relevant. If a system produces a result that contains a recognizable part of a protected work, such output may legally qualify as reproduction or even as communication to the public where the content is publicly accessible.
Emerging case law
The development of technology is already being accompanied by the first significant court proceedings. Among them are disputes concerning the use of copyrighted works in training generative AI models.
In Europe, particular attention has been drawn to the GEMA v OpenAI dispute, in which the German collective management organization alleged that lyrics from its repertoire had been used to train language models and that a chatbot subsequently returned recognizable parts of those lyrics in response to user prompts. Among other issues, the court examined whether so-called memorisation—the retention of text in a model in a manner that allows it to be reproduced—could constitute reproduction within the meaning of copyright law.
Case law relating to authorship in AI-generated works is also evolving. In a dispute concerning graphic logos created with the assistance of generative artificial intelligence, the court emphasized that general prompts to the system or the mere selection between multiple generated outputs will usually not suffice for copyright protection. For copyright protection to arise, the human creative influence must be clear and predominant in the final result.
Important proceedings are also ongoing in the United Kingdom and the United States, such as the dispute Getty Images v Stability AI, which concerns the use of millions of photographs to train generative image models. In such cases, courts frequently begin by examining the technical functioning of the models, as understanding the distinction between the training phase and the generation phase is crucial for legal assessment.
Does the existing legal framework suffice?
A key question for the future of copyright law is whether the existing legal framework—particularly the European regime based on the InfoSoc Directive and the DSM Directive—remains sufficient to regulate relationships in the age of generative artificial intelligence.
On the one hand, the core concepts of copyright law still appear applicable: the right of reproduction, the right of communication to the public and the concept of the author’s own intellectual creation continue to form the basis of legal analysis. On the other hand, generative AI raises new questions that legislation has not yet explicitly addressed, particularly regarding model training, liability for generated outputs and the boundary between human and algorithmic creativity.
For this reason, discussions increasingly arise as to whether certain aspects of artificial intelligence may require a new sui generis legal regime that would specifically regulate the relationship between AI systems, data and copyright protection.
AIxPravo: discussing the future of law and technology
These issues were also presented by partner Eva Gostiša and attorney Nejc Setnikar at the AIxPravo conference, where our firm participated both as speakers and as the authors of the challenge prepared for students participating in the legal hackathon. The discussion at the event highlighted how rapidly the field of AI law is developing and how important it is to connect legal expertise with technological understanding.
For our firm, it was also interesting to observe that we are currently the only law firm in Slovenia that have established a dedicated role of Head of Legal Technologies and Innovation. This role is held by Aljaž Jadek, an experienced attorney who has followed technological developments for decades and leads the development of legal technologies within our firm.
