Politics and regulation·November 3, 2026, 10:53

EU adviser: Chatbot providers may be liable in some cases for reproducing articles

AI-generated from the sources listed, with automated source checks.

An advocate general at the EU Court of Justice says chatbot providers can, under certain circumstances, be held liable if their responses recognizably reproduce the form of a news article. The case concerns Google's Gemini and a Hungarian publisher.

An advocate general at the EU Court of Justice has issued an opinion in a case on whether Google's chatbot Gemini infringes copyright when it uses content from news outlets. According to the opinion, chatbot providers can in some cases be held liable. It is not a general conclusion that all use of news is an infringement.

The case was brought by the Hungarian publisher Like Company. The publisher claims that Gemini reproduced its articles between June 2023 and May 2024.

Form, not facts

Copyright protects the specific form of expression and the investment behind the publications. It does not protect facts or news as such. That means a chatbot can tell users what has happened without that in itself being a problem.

According to the opinion, however, it can become a problem for the provider if a response recognizably reflects the form of the article, so that the user does not need to read the original. The advocate general also looks at whether the chatbot gained unlawful access to the content, for example by bypassing a paywall. Separately, the provider can be held responsible if the reuse is part of its commercial strategy, or if it has not taken credible and effective measures to prevent it.

Training the model

The case also raises questions about whether articles were used to train Gemini. Here, the advocate general recommends that the questions be dismissed as inadmissible. The reasoning is that there is no evidence that Like Company's articles were used to train the model. As a comment on the substance of the case, even though the questions on training are recommended to be dismissed, the advocate general takes the view that using protected material for training generally falls under the exception for text and data mining.

What does it mean?

An advocate general's opinion is not a ruling. It is a recommendation to the judges, who often, but not always, follow it. It is still unknown what the court will conclude and when the ruling will come.

For readers who use chatbots to stay informed, the opinion suggests that the question of how much a chatbot may reproduce from an article is being settled in EU law. For news outlets, it could affect how their rights apply to AI services.