ECJ opinion: AI chatbots can violate press copyright

dpa | According to an opinion at the European Court of Justice (ECJ), AI chatbots like Google's Gemini can violate copyright. If the chatbots retrieve press articles from websites and use them to provide users with an answer, this could affect the rights of publishers to reproduce and make publicly available, said Advocate General Maciej Szpunar in his opinion, according to the Court's announcement.

The much-noticed case is the first on generative artificial intelligence before the ECJ in Luxembourg. The report is not binding for the judges; their verdict should be announced at a later date.

According to the Advocate General, whether there has been a duplication depends, on the one hand, on whether the chatbot accessed the article unlawfully, for example by circumventing technical measures such as a payment barrier. On the other hand, Szpunar emphasizes that the law of press publishers does not protect the reported facts, but only the “specific form of expression” of the publications. The law should not give publishers a monopoly on news.

The decisive factor is whether the answer reuses the form of expression in a recognizable way, in such a way that readers no longer have any reason to call up the original text. The right of public access is generally violated if the form of expression in the answers is “systematically recognizable”.

The background is a legal dispute in Hungary. The press publisher Like Company, which operates several news portals, sued Google there. The accusation: Gemini reproduced the publisher's protected texts without permission and made them publicly accessible. The responsible court in Budapest submitted questions about EU law to the ECJ. Ultimately, the case must be decided by the Hungarian court, which is bound by the requirements of the ECJ.

The Advocate General considers two questions from the Hungarian court regarding the training of AI models to be inadmissible. They are hypothetical because there is no evidence that articles from the plaintiff publisher were used to train the Gemini models. However, he noted that the use of protected works to train an AI model is “freely permitted” under a legal exception known as the text and data mining barrier. Under certain conditions, it allows texts and data to be automatically read and analyzed in order to recognize certain patterns.

A Google spokesperson said the company welcomed the statement on AI training. “Gemini learns linguistic patterns to generate original content rather than reproducing content,” it said. Rights holders would also have the opportunity to object to the use of their content.