Share This Article
The CJEU is set to clarify how EU copyright rules apply to generative AI – a legal turning point for the future of artificial intelligence in Europe.
In the fast-developing intersection of AI and copyright, the Court of Justice of the European Union (CJEU) is now preparing to issue its first ruling on how copyright law applies to generative AI technologies. The preliminary reference in Like Company v Google Ireland (C-250/25), originating from Hungary, raises fundamental questions about the copyright implications of AI-generated content, including whether AI training or outputs infringe reproduction or communication rights under EU law.
Background: A Copyright Dispute Involving AI and a Press Publication
The case stems from an article published on the Hungarian website balatonkornyeke.hu, owned by Like Company. The article, reporting on singer Kozso and his unusual plans involving dolphins, was compiled from publicly available online sources and included a Facebook image.
When a user prompted Google’s Gemini chatbot to summarize this article in Hungarian, the resulting output allegedly replicated the core information of the original piece. Like Company argued that the chatbot’s summary constituted unauthorized use of its copyrighted content, violating its press publisher rights under Article 15 of the Digital Single Market (DSM) Directive – which protects digital uses of press publications.
Google’s Response: No Copyright Infringement by AI
Google rejected the infringement claim, raising several legal and factual defenses:
- No applicable copyright law: Since the output wasn’t generated or stored in Hungary, Hungarian copyright law would not apply.
- No new public reached: The article was freely accessible online, so summarizing it did not qualify as “communication to the public.”
- No verbatim reproduction: The chatbot’s summary didn’t reproduce protected expression, but only presented factual content.
- Applicable exceptions: Even if reproduction occurred, exceptions under EU law applied—such as:
- The temporary copies exception (Article 5(1), InfoSoc Directive)
- The text and data mining (TDM) exception (Article 4, DSM Directive), which permits AI tools to extract information from lawfully accessible content unless opted out by the rightsholder.
The CJEU’s Involvement: Preliminary Questions on AI and Copyright
The Budapest Környéki Törvényszék (District Court) referred several questions to the CJEU, including:
- Does a chatbot’s AI-generated response that replicates content from a protected publication constitute a reproduction or communication to the public, even if the content is generated via prediction?
- Does the training of AI models on copyright-protected content trigger the reproduction right under EU law?
- If so, can such training fall under the TDM exception in the DSM Directive?
These questions place copyright, AI, and the CJEU at the center of a critical legal debate with wide-reaching consequences.
Key Legal Considerations on AI and Copyright
Two points are essential before diving deeper:
- Protection status of the content: The Hungarian court appears to assume that the article is protected by copyright. However, EU copyright law doesn’t cover mere facts, and news compilations may only qualify for protection if they exhibit sufficient originality.
- Independent creation and derivation: Even if content is protected, infringement requires a connection between the AI output and the original work. Establishing that a language model derived its output from a specific source is difficult due to the opacity of training data, a challenge currently addressed by transparency obligations in the upcoming AI Act.
If both protection and derivation are established, the CJEU is likely to follow its tradition of strong IP protection (Article 17(2), EU Charter) and rule that even AI-generated predictions may infringe copyright if they replicate substantial protected content.
AI Training and the Reproduction Right under EU Law
One of the most contentious issues in the copyright-AI debate is whether training AI on protected works amounts to reproduction. Under EU copyright law, the answer is generally yes.
The InfoSoc Directive defines reproduction broadly. The TDM exceptions in the DSM Directive were introduced precisely because AI training (or any automated extraction of content) would otherwise infringe this right. Denying that training triggers reproduction would nullify the purpose of these exceptions.
Applicability of the TDM Exception to AI Training
Can AI developers rely on the TDM exception for training purposes? The DSM Directive does allow the use of protected content for TDM purposes under certain conditions:
- The content must be lawfully accessible.
- Rightsholders can opt-out.
- The use must meet the three-step test: it must serve a special purpose, not interfere with normal exploitation, and not unreasonably harm rights holders.
While some commentators initially questioned whether the TDM exception covers AI model training, EU institutions have since clarified that it does—especially with the new AI Act explicitly referencing this link. However, the exception applies only to certain acts (like reproduction), not to AI training as a whole.
What This Means for the Future of AI and Copyright Law
A final decision from the CJEU is not expected before late 2026 or early 2027. Still, this preliminary reference is a landmark case. It will define how copyright law applies to AI, including both training and output generation, and clarify the relationship between press publisher rights, generative AI, and text and data mining exceptions.
As AI becomes more integrated into creative and publishing industries, this decision could reshape the regulatory framework across Europe—redefining the boundaries between innovation and intellectual property protection.
On the topic, you can read the article “AI Act – What Is the Scope of the TDM Copyright Exception?“. Also, you can read the latest issues of our DLA Piper’s AI law journal HERE.