Preliminary questions to the Court of Justice on AI, chatbots and copyright

The rise of AI systems such as chatbots increasingly raises legal questions, especially when it comes to training AI systems with copyrighted material. Recently, a number of preliminary questions have been referred to the Court of Justice of the European Union ('Court') in this regard.

What exactly is going on? What preliminary questions have been referred to the Court? And why are the answers of great importance to anyone using or developing AI systems?

What are the relevant facts?

The relevant parties in these proceedings are Like Company ('Like'), a Hungarian publisher of news websites, and Google Ireland ('Google'). Like discovered that Google's AI chatbot (Gemini, formerly Bard) was able to provide summaries of articles published on Like's website at users' request. Sometimes even detailed excerpts of these articles were displayed, including source citations.

Gemini's chatbot was trained to use large amounts of text to recognise patterns in order to generate responses. In doing so, Gemini uses the articles as published on Like's news websites. Like believes that Google is thereby infringing its copyright because its copyrighted news articles are reproduced by Gemini without Like's prior consent and made available to the public through the chatbot.

Because Like is relying on European laws and regulations (Directive 2019/790 on copyright and related rights) in this matter, the Hungarian court can ask the Court precise questions (preliminary rulings) that should interpret these laws and regulations in this specific context. The Court can thus provide answers that are relevant to this particular case, but in addition have relevance for the other EU member states.

What preliminary questions have been asked?

The Hungarian court asked the Court for clarification on four key issues:

1. Is the display of text by an AI chatbot a 'communication to the public'?

Should the fact that a chatbot displays text from news articles be seen as a copyright-relevant act requiring permission from the copyright holder, in this case Like?

2. Is training an AI model based on existing text a 'reproduction'?

Does analysing and processing texts while training a chatbot fall under the copyright concept of 'reproduction'.

3. Does this reproduction fall under the text and data mining exception?

If training an AI model is indeed a reproduction, can it be done without the copyright holder's permission based on the European rules for text and data mining?

The exception for text and data mining from Article 4 of Directive 2019/790 on copyright and related rights allows the reproduction of copyrighted works for text and data mining without the copyright holder's permission, provided those works are lawfully accessible and the use meets the legal conditions. The question is whether the use of Gemini's news articles can be brought under this exception and/or it was therefore necessary for Google to seek permission from Like before using the news article in Gemini.

4. Is the generation of a response by the chatbot a reproduction by the provider?

If a user issues a command to Gemini that leads to the display of (parts of) a press publication, is Google as provider of Gemini responsible for this reproduction?

Conclusion

The case at the Court is a milestone for European copyright law in the AI era. For instance, it will clarify whether providers of AI systems have to seek permission from copyright holders such as publishers to present summaries or excerpts of news articles. It will also answer the question of whether training AI systems without author's permission is possible and how the text and data mining exception plays a role in this. Finally, the question of who is responsible for the output of an AI system will also be clarified. Is this the user or the provider or both?

So this case is important not only for big tech like Google, but also for start-ups, publishers, and basically anyone who deploys AI for processing text. The outcome will determine in what way AI systems can use copyrighted material and where the limits of copyright lie in the digital age.

More information.

Once the Court provides the answers to the preliminary ruling questions, we will post about it. In the meantime, should you have any questions on the legal aspects of AI, our specialists are at your disposal.