Recently, the Budapest Environs Regional Court (Budapest Környéki Törvényszék) in Hungary referred a case to the Court of Justice of the European Union (CJEU) concerning the use of press content by generative AI. This is the first time the CJEU has been asked to clarify how EU copyright law applies to the training and functioning of large language models (LLMs).
Facts of the case
The first AI copyright case, Like Company v. Google Ireland (C-250/25), was brought by a Hungarian press publisher against Google’s chatbot Gemini (formerly Bard). The publisher argues that Gemini reproduced and summarised its protected articles when prompted by users, including a report concerning a Hungarian celebrity. According to the publisher, this conduct undermines its advertising-based business model, since readers may rely on the chatbot’s output rather than visiting its websites.
According to the referral, the Hungarian court seeks guidance on four main questions:
- Whether displaying chatbot responses that reproduce substantial parts of press articles amounts to “communication to the public,” and whether it matters that the text is generated by predictive modelling.
- Whether training an LLM on protected works should be regarded as “reproduction” under copyright law.
- If so, whether such training can be justified under the text and data mining exception.
- Whether a chatbot provider is responsible when the system generates outputs that include protected content from press publications.
The publisher maintains that both the training and the operation of the chatbot go beyond what is permitted by law. In particular, it emphasises that the exception for “very short extracts” in Article 15 of the CDSM Directive does not apply, as the summaries generated by Gemini often contain substantial parts of its original works. Google, however, denies infringement, explaining that Gemini does not store or retrieve articles, but generates responses by modelling patterns in language. It also highlights that the resemblance between generated text and original works may be coincidental or a result of “hallucination,” a known characteristic of such systems. Based on this, the relevant EU copyright exceptions for temporary copies and text and data mining apply.
Significance
The significance of this case is considerable. For the first time, the CJEU will need to decide how far existing copyright rules extend to generative AI systems. If the Court takes the publisher’s position, AI developers may be required to obtain licences before training their models on press content or risk liability for outputs. This would substantially change current practices and strengthen the bargaining power of rights holders. If, however, the Court accepts Google’s arguments, developers may continue to rely on publicly available material for training without specific licences, keeping barriers to innovation relatively low.
Beyond this dispute, the ruling will clarify the scope of the text and data mining exception, which has become central to the business models of many AI companies. It will also help define whether and to what extent AI providers can be held liable for the content produced by their systems.
Practical Takeaways
Although the ruling is still pending, the case shows that:
- Courts and regulators are beginning to test how existing copyright rules apply to AI.
- Businesses using or developing AI tools should be prepared for stricter compliance requirements in the future.
- Press publishers and other rights holders may gain stronger grounds to demand licences or negotiate compensation for the use of their content in AI training.
The judgment of the CJEU will therefore be closely watched across the technology and publishing sectors and is likely to become a reference point in shaping the balance between innovation and copyright protection in the years ahead.
