A Hungarian case before the Court of Justice of the European Union asks how generative artificial intelligence should be allowed to reuse online news content. In his written opinion, Advocate General Maciej Szpunar recommends that the Court acknowledge that, in certain circumstances, generative AI chatbots may infringe press publishers’ rights when they retrieve online newspaper articles and reuse those articles’ content in responses. The dispute was brought by Like Company against Google Ireland and was prompted by a controversy over a piece related to plans to introduce dolphins to Lake Balaton that referenced the entertainer Kozsó.
Key points of the Advocate General’s opinion
Szpunar’s opinion sets out several distinguishing principles and tests:
- Facts versus expression: copyright-related protection for publishers does not give them exclusive rights to the facts reported in news articles, but it does protect the distinctive expression and the investments publishers make to produce content.
- Two-part test for potential infringement: to determine whether a chatbot’s answer violates a publisher’s reproduction right, one must examine (1) whether the system obtained the original article lawfully, with particular attention to whether it bypassed effective technical measures that restrict access or automated downloading (for example, a paywall), and (2) the extent to which the chatbot’s reply reproduces the original publication’s distinctive mode of expression. Merely summarizing factual content does not automatically constitute infringement; however, if the output copies recognizable expressive features to such a degree that users no longer need to consult the original article, that may infringe the publisher’s rights.
- Public access right: the right of making works available to the public may be infringed if a chatbot repeatedly and recognizably reproduces publishers’ distinctive expressions in responses to related queries.
Training data issues
The Hungarian court also asked the EU Court about the legal classification of content used to train AI models, not only about the chatbots’ answers. Szpunar recommends that the Court declare those questions inadmissible. He reasons there is no evidence that Like Company’s articles were actually used to train Google Gemini’s language models, and that training language models, by itself, does not amount to online use of press publications.
At the same time, the Advocate General notes that using protected content to train language models can, in principle, be permitted under the EU exception for text and data mining.
Liability of the service provider
Szpunar further states that a chatbot provider can be held liable for such uses if the practice forms an integral part of its business model or if the provider fails to take credible and effective measures against infringing reuse of content.
Why the case matters
The dispute’s significance goes well beyond a single Hungarian news item. The Court’s interpretation could set rules on how generative AI services may access and repurpose online press content and under what limits. If chatbots routinely produce answers that sufficiently substitute for reading original articles, publishers may lose traffic and part of their business models. Because the procedure is an Article 267 preliminary reference, the Court’s interpretation will guide not only the underlying national dispute but also other national courts facing similar questions across EU member states.
Legal process and next steps
An Advocate General’s opinion is influential but not binding on the Court’s judges; the judges are now deliberating and will issue a ruling in due course. In practice, the Court often follows the Advocate General’s reasoning, but its final judgment may differ. The case remains a central legal test at the intersection of copyright, digital markets and AI regulation in the EU.
