Introduction

In a move likely to shape the future of Artificial Intelligence (“AI”) and copyright law across the European Union, the Court of Justice of the European Union (“CJEU”) has received its first referral concerning AI generated outputs and press publishers’ rights. The referral, Case C-250/25 Like Company v. Google Ireland, refers to the court to determine whether the reproduction and public display of press content by a large language model (“LLM”)  infringes copyright and related rights law.

What is at stake here?

The outcome of this case may reshape the digital publishing and AI sectors’ understanding of lawful use, the implications of text and data mining and the scope of press publishers’ rights under the DSM Directive given this context.

Background

The case originated in Hungary. Like Company, a local press publisher operating various news portals, claims that Google’s chatbot Gemini reproduced and summarised parts of its protected journalistic content without consent.

The publisher Like Company argues that Google’s generative AI services unlawfully reproduced and made its content available to the public. According to the claim, these outputs went beyond “very short extracts” and thus triggered the press publishers’ related rights granted under Article 15 of the DSM Directive, and reproduction rights under Article 2 of the InfoSoc Directive.

Google delved into the technicalities and the functionality of the underlying architecture, and countered that the chatbot is not an information database, nor an information retrieval system. Google adds that it does not store copies of the data collected but rather converts them into tokens and processes them for the purposes of predicting text based on patterns. Therefore, it argued, there is no unlawful reproduction or communication to a “new public”.

The questions

The Budapest Környéki Törvényszék has asked the CJEU to clarify:

1. Whether chatbot outputs that reproduce parts of press articles qualify as acts of reproduction and communication to the public under the DSM and InfoSoc Directives even if the AI is just predicting based on patterns.
2. Whether LLM training constitutes reproduction under Article 2 of the InfoSoc Directive.
3. If so, does such reproduction qualify for the TDM exception in Article 4 of the DSM Directive?
4. Whether user-prompted generation of content matching protected works amounts to actionable reproduction by the service provider.
   
   

Reproduction, exceptions, and responsibility

Whilst this area is relatively nascent, current legal literature differentiates between training and output.

Training often involves large-scale reproduction of protected material. While the TDM exceptions under Articles 3 and 4 of the DSM Directive allow limited acts of extraction and reproduction, they do not permit downstream acts like generation of outputs that mirror the training data. It is to be noted that the AI Act reinforces that even training conducted outside the EU must comply with EU law if the model is marketed within the Union.

On the other hand, output generation, where AI produces content resembling, using or quoting copyrighted material, raises further issues and may lead to copyright infringement.  

Liability

A landmark aspect of this case is the potential for AI providers to be held liable for infringing outputs. EU and UK case law suggest that liability may extend beyond users to developers and providers of models, especially if they play an “indispensable role” in the act of communication or fail to implement safeguards to prevent unlawful use.

This view echoes the logic behind platform liability rulings in YouTube,[1] Ziggo,[2] and TuneIn,[3] where service providers were found liable for facilitating or not preventing infringement. Contractual disclaimers in terms of service may not shield providers from liability if courts find they had knowledge of or enabled infringing uses.[4]

What’s Next?

The CJEU’s ruling, expected no earlier than late 2026, could have sweeping implications. It will clarify whether chatbot responses can lawfully include verbatim or near-verbatim reproductions of protected text, under what conditions AI training is permissible, and how the press publishers’ right interacts with evolving AI capabilities.

In the meantime, one thing is certain; the lines between fair use, training, and reproduction are being redrawn in real time. For press publishers, AI developers, and policymakers alike, this referral may be the first major test of where copyright ends, and where algorithmic authorship begins.

--

GTG will assist you in planning your AI compliance. Do not hesitate to contact us on info@gtg.com.mt for any information.

Author: Dr J.J. Galea

Continue reading on:

Banned AI: What the EU AI Act Means for Gaming and Gambling Systems

AI, the uninvited guest

Clarifying Obligations under the AI Act: A Preview at the Commission’s Targeted Consultation on General-Purpose AI Models

[1] C-682/18 / C-683/18

[2] C-610/15

[3] EWCA Civ 441

[4] E. Rosati, Infringing AI: Liability for AI-Generated Outputs under International, EU, and UK Copyright, Law European Journal of Risk Regulation, 1–25. https://doi.org/10.1017/err.2024.72