The Court of Justice of the European Union (CJEU) has been referred a case, namely Like Company v Google Ireland C-250/25, that may shape future EU copyright law on artificial intelligence (AI). In short, the Budapest Környéki Törvényszék, i.e. the Budapest District Court in Hungary, has asked the CJEU to decide whether the training and functioning of a large language model (“LLM”), which understands, processes, and generates human language responses in AI, can constitute reproduction and/or communication to the public under EU copyright law.
For background, Like Company, a Hungarian news publisher, alleged that Google’s AI chatbot named Gemini reproduced and made copies of its news articles, that are protected by copyright law, without consent. Gemini can produce a detailed summary of an article for a user, which Like Company claim is a reproduction of its substantial content. The article used in this case was Like Company’s report of a Hungarian singer’s plans to introduce dolphins to Lake Balaton.
The referring court has asked the CJEU to interpret EU copyright law, specifically the 2001 Copyright and Information Society Directive (InfoSoc Directive), and the 2019 Copyright in the Digital Single Market Directive (CDSM Directive), in the context of AI and decide on the following: (1) whether reproducing content in the chatbot response that is contained within the article exceeds a “very short extract” that is permitted to be communicated to the public, (2) whether the development of an LLM constitutes reproduction given that it uses predictable formations in a language and substitutes sensitive data from protected works, (3) if said AI training constitutes reproduction, whether it would qualify for the text and data mining exception that allows researchers to use copyrighted material for non-commercial research purposes, provided the access to the materials is lawful (which would be determined by question (2)), and finally (4) whether the responses generated by chatbots, at the request of a user, qualify as reproduction by the provider of AI (here Google) and thus would require consent from the copyright holder in respect of such responses.
Like Company argues that the responses produced by chatbot exceeds the permitted “very short extract”, and therefore surpasses what it permitted to be used lawfully without permission. In turn, the publisher alleges that the responses should not be held to qualify for the text and data mining exception, as the purpose of training the LLM does not constitute valid research. Further, Like Company claims that the responses produced by the chatbots take away visits from websites that has cost owners substantial revenue in advertising. For example, website owners can pay extra in sponsored keywords and advertisements on Google and for their websites to appear in the top search results, but now the first result is simply replaced by the AI generated response which pushes the website further down in the appearance of the search results and thus is deferred.
In its defence, Google’s case is that rather than hoarding copies of articles to reproduce, Gemini is trained to break down data and relies on AI to produce text based on probabilities or likelihoods, which suggests that it determines the suitability of the generated response. Therefore, it is reliant on the exceptions provided by EU law for temporary acts of reproduction in generating responses, and also the text and data mining exception. Google asserts that Gemini is a supportive tool rather than a collection of data, and because its users already have access to the websites that contain the original content that it derives data from it does not reach any new audience, and thus does not take away website visits and does not cause any damage to website owners.
The referral is just the first step in determining the issues as the decision is likely to be issued in 2027, or late 2026 at the earliest. Should the CJEU decide the case in favour of Like Company, in that the training of AI and its generated responses exceed what is permitted in the EU exceptions, developers may be required to obtain permission and licences from copyright holders before using their content to train its AI system and reproducing their content, which would be positive for copyright holders enforcing their legal rights, and may enable them to put a price on using their protected content, but would be a costly legal slippery slope for AI developers. If the CJEU take the alternative approach, AI developers may continue to use protected content without consent of the holder and copyright holders such as Like Company would have no choice but to allow it, unless they blocked access to data which may also block the public. The CJEU’s decision is expected to clarify the extent what is considered appropriate to train AI systems pursuant to the exceptions, which may impose significant impacts on the way AI is developed and trained in respect of interpreting and using protected content. Either way, the outcome will be a landmark decision for the future development and regulation of AI in Europe.
If you have any questions on the above, please do not hesitate to contact the team at McDaniels Law on 0191 281 4000 or legal@mcdanielslaw.com.
