Copyright vs AI in the Court of Justice of the EU

By Johan Åberg

Of course, it was only a matter of time before AI, artificial intelligence, would end up in a formal dispute with copyright and need to be scrutinised by the European Court of Justice. Admittedly, this is not the first time that new technologies have interfered with the purpose and principles of copyright law; in the past, it has been satellite decoders (C-403/08), hyperlinks (C466/12, etc.) and file-sharing platforms (C-610/15). But this time it is a more revolutionary technology that is likely to change our view of intellectual creation and its value.

One of the areas of conflict arises from the fact that AI, in order to achieve its functionality and full potential, must train on "MI", human intelligence, i.e. the AI service must read and encode texts, images and music that have been created by humans. As a starting point, these texts, images and music are protected by copyright. The legally interesting - and economically significant - question is therefore whether AI services that copy texts, images and music without authorisation in their training infringe exclusive copyright or related rights. This question has now been referred to the Court of Justice for the first time (C-250/25).

In April 2025, the Budapest District Court asked the Court of Justice of the European Union for a preliminary ruling in a case between Like Company Kft, a Hungarian news organisation, and Google Ireland, the provider of the chatbot Gemini. The Hungarian news company claims that Gemini reproduced one of its news articles without authorisation, which was discovered when a user asked Gemini to summarise the news article in question. Gemini generated a detailed response which, according to the Like Company, reproduced parts of the original article, which would constitute unauthorised copying and communication to the public giving rise to a right to compensation.

The rights which the Court of Justice of the European Union will examine in this case are thus those which newspaper companies (referred to in law as 'producers of press publications') acquired in 2019 under Article 15 of the Digital Single Market Directive, which was implemented in Sweden on 1 January 2023 in Paragraphs 48b-d of the Copyright Act. This neighbouring right allows newspaper companies to control the online use of their publications, which means that, subject to certain limitations, they can claim remuneration for the use of their material online.

As counsel in a couple of the Swedish copyright preliminary rulings from the CJEU ("Svensson" C-466/12 and "Stim and SAMI" C-753/18), I note that Google's argumentation raises some of the questions of principle on which the CJEU has already ruled. Google argues that the same users could access the article in question directly, which means that no "new public" is reached, but also that Gemini's response only constitutes single words and very short extracts from the original article. Furthermore, Google argues that all uses fall within the legal exceptions for temporary reproduction and text and data mining, and that Article 15 is based on a balancing exercise where the aim of compensating newspaper companies for their investment is not as important as protecting users' freedom of expression and information.

How the court will assess Google's challenge in the light of existing law will have a commercial impact on rights holders and AI service providers, either in the form of demands for a licence and compensation, or that rights holders will have to accept that the AI services fall outside the copyright framework. Given that it takes 12–18 months from the date the request for a preliminary ruling is received by the Court of Justice of the European Union until a judgment is delivered, it is likely that a judgment in C-250/25 will be delivered in the autumn of 2026. Whatever the outcome, it is probable that the party that draws the short straw will want to see a change in legislation.

Here is a summary of the Hungarian court's request for a preliminary ruling.

The position of Swedish right holders on the use of copyright material by AI services for training purposes is uniform and consistent, namely that it requires a licence, see for example

In this context, it is worth noting that the Swedish Agency for Digital Government (Digg) touches on the issue in its guidelines on generative AI in public administration but notes that "Exactly how the training of AI should be regarded legally is not clear."

In the absence of clear rules, the market parties have so far acted somewhat differently. In February 2025, for example, Schibsted entered into an agreement with OpenAI, which the British Financial Times also did the year before, meaning that OpenAI can "train" on Financial Times articles. However, several American media houses have acted differently and sued OpenAI and Microsoft for copyright infringement, including the New York Times. Other rights holder groups have also chosen to let the courts judge the use of copyright works by AI services, for example GEMA (the German equivalent of Stim) has sued both SunoAI and OpenAI in court in Munich.

In conclusion, given the enormous importance of copyright, economically and culturally, in light of the rapid development and use of AI services, the court decisions, negotiations and agreements of the next few years will be decisive for the conditions of future intellectual creation, regardless of whether the creator is abbreviated AI or MI.