AI Chatbots and copyright: Anticipating the ruling in CJEU’s Case C-250/25 Is there any space left for generative AI in the EU copyright legislative framework?

Like Company v Google Ireland Limited marks the first referral to the Court of Justice of the European Union (CJEU) for a preliminary ruling concerning generative AI (GenAI) and copyright. Referred by the Budapest District Court in spring 2025, Case C-250/25 seeks to assess whether Google’s Gemini (LLM-based chatbot) infringed press publishers’ rights by outputting text reproducing (parts of) press articles available on a publisher’s website. Rasing multiple questions regarding both the Directive 2001/29/EC (InfoSoc) and the Directive (EU) 2019/790 (DSM), this case offers the CJEU a key opportunity to clarify whether and under what conditions AI training and AI-generated outputs fall within the scope of copyright protection.

ANALYSIS OF THE QUESTIONS REFERRED

The questions referred to the Court are, in essence, the following:

1.      Does the display, in the responses of an LLM-based chatbot, of a text partially identical to the content of web pages of press publishers, where the length of that text is such that it is already protected under Article 15 of Directive 2019/790, constitutes an instance of communication to the public? If the answer to that question is in the affirmative, does the fact that [the responses in question are] the result of a process in which the chatbot merely predicts the next word on the basis of observed patterns have any relevance?

Our view: market voices claim that GenAI practices and services could pose a new challenge to the survival and operation of free press, as it is capable of assembling content based on input data that mimics human creativity. According to Article 3(1) of the EU Artificial Intelligence Act, an 'AI system' is defined, in essence, as a machine-based system designed to operate with varying levels of autonomy and which infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments.

Under Art. 2 InfoSoc, as a general rule, a copy of a protected work equals reproduction: if the chatbot’s answer is identical to (a substantial part of) a publisher’s text, it prima facie involves copying, and could be infringing upon the authors’ rights to authorize the reproduction and the making available to the public of the work. Also, article [AST1] 15 DSM grants the publisher the exclusive right to authorize such reproduction by information society services.

In the case law, “communication to the public requires making content available to a public beyond the one initially targeted”.

The chatbot answer is directed to the user, but that user is accessing a copy of the publisher’s article. Unlike in the case of the use of a hyperlink, the chatbot delivers the text itself. It could be held that the authors did not consider such use of their works at the time of the publishing. Hence, it could be considered that the chatbot delivers the content to a new audience.

If one of the functionalities of the chatbot service is to summarize and partially communicate works (press publications), it could be argued that this could significantly disrupt the normal ways of exploitation by the press publishers themselves and their economic interests, which are usually fulfilled directly through their websites and by the advertising made on their websites.

2.      Does the process of training an LLM-based chatbot constitute an instance of reproduction, where that LLM is built on the basis of the observation and matching of patterns, making it possible for the model to learn to recognize linguistic patterns?

Training involves processing large amounts of text, by extracting and reproducing information and patterns, using even press publications (which could be copyrighted works), often scraped from online sources, without any permission or compensation. If copyright is found in this input, it may also subsist in the output produced by the model. Therefore, it could fit the broad definition of reproduction under the EU law: any direct or indirect, temporary or permanent copying by any means.

An important question to be answered is what happens if AI chatbots internalize protected work, generating output using copyright protected input. Could it be held that the mere internalization for the purpose of a future use of that input constitutes a violation of the reproduction right? A positive answer could be supported by the CJEU judgment issued in case Infopaq, where even temporary or invisible reproductions were found to infringe Art. 2 of InfoSoc.

3.      If the answer to the second question referred is in the affirmative, does such reproduction of lawfully accessible works fall within the exception provided for in Article 4 of Directive 2019/790, which ensures free use for the purposes of text and data mining?

This question is the most complex and decisive for the issue and we expect that the Court’s interpretation will be nuanced.

Text and data mining (TDM) is relevant to AI training but is not solely what AI training consists of. While both TDM and GenAI rely on large-scale data, they work in fundamentally different ways: TDM finds patterns, while GenAI synthesizes new expressions. Therefore, if the TDM could be covered by certain exceptions, it is not clear if the GenAI falls under the same/a similar legal treatment.

If it does, Art. 4 of DSM could be applied. In fact, the EU’s 2024 AI Act could be interpreted as implicitly endorsing that AI training falls under Article 4, provided that Art. 4 (3) of DSM is observed. Thus, all else equal, the reproduction during training could be lawful under Art. 4 if (for example) the publisher’s works were accessed from the open web and no technical TDM block was signaled.

If the training could be exempted under Art. 4 DSM under certain circumstances, it would, however, be difficult to argue that the final answer of the chatbot would also be covered by this exception. The conditions under Art. 4 are the following: lawful access, no reservation by the rightholder and that the copies are made solely for TDM purposes. Therefore, the Court shall analyze if the AI training meets these criteria.

However, it could be argued that even if the exception of Art. 4 would protect the input stage, it would not be able to protect the output stage and at least that shall be considered an infringement under Art. 4. Moreover, under settled case law from the CJEU (Infopaq, Pelham), exceptions to copyright must be interpreted strictly.

As such, it remains to be seen whether GenAI fits into the definition of text and data mining (TDM). The legal definition of TDM in Arts. 3 and 4 of DSM limits its scope to analytical, generative uses. GenAI operation mode seems, however, to be trespassing the function of TDM, aiming to create new output instead of merely identifying trends, patterns and correlations based on the analysed data. Hence, it is left to see if such practices fall outside the intended scope of Arts. 3 and 4 of the DSM.

4.      Must Article 15(1) of Directive 2019/790 and Article 2 of Directive 2001/29 be interpreted as meaning that, where a user gives an LLM-based chatbot an instruction which matches the text contained in a press publication, or which refers to that text, and the chatbot then generates its response based on the instruction given by the user, the fact that, in that response, part or all of the content of a press publication is displayed constitutes an instance of reproduction on the part of the chatbot service provider?

From our perspective, we would rewrite this as: should the chatbot service provider be held liable for an output generated based on the user’s instructions?

The question is very complex, given the multitude of situations that may arise in practice. We could imagine, for example, a situation in which the user, by the prompt provided to the chatbot, sought to obtain a response including an actual reproduction of a protected work. In such a case, holding the sole responsibility of the service provider could seem unfair. 

However, the fact that the user is providing a prompt to the chatbot should not make it solely responsible for the result generated by the chatbot. It could be held that the facts should be analysed separately—input on the one hand, and output on the other, to see if unauthorized reproduction can be found in each.

Nevertheless, we believe that analysis in such situations should be carried out on a case-by-case basis.

RULING SCENARIO ANALYSIS AND CONSEQUENCES

By analyzing the context, we believe that it is plausible that the CJEU will adopt a rights-protective stance, in line with the EU’s consistent position to enforce copyright protection.

IF THE COURT RULES IN FAVOR OF PUBLISHERS:

Chatbot service providers may be forced to license news content to include it in outputs and policymakers might be urged to clarify Art. 4 DSM (perhaps expanding its scope or adding specific research exemptions for AI) or to revise Art. 15, in order to, for example allow for “transformative” uses for AI.  

IF THE COURT RULES IN FAVOR OF AI:

For the press industry, this could be a setback, because publishers could lose traffic and ads revenue as more users could rely on chatbots for accessing press articles instead of accessing the publishers’ websites directly. 

PERSPECTIVES

Case C‑250/25 marks a turning point in European copyright legislative framework. This ruling is poised to become a cornerstone in shaping how the EU balances copyright with technological progress and its full implications will unfold in the years to come, demanding vigilance from all stakeholders.