EU Court Adviser Says Chatbot Answers Can Breach Press Publishers' Rights, but Training Is a Separate Question

By Dafa Attas

Photo: jestemrobert / Pixabay

The EU's top court has received its first advisory opinion in a dispute about generative AI. Advocate General Maciej Szpunar gave his view on 3 November 2026 in Like Company v Google Ireland (Case C-250/25), a reference from Hungary about whether answers produced by Google's Gemini chatbot (formerly Bard) can infringe the special right that EU law gives to press publishers.

The opinion does not decide the case. The judges will rule later and are free to depart from it. Still, opinions of this kind often shape the final judgment, and this one gives the first detailed map of how EU copyright rules might apply to AI assistants that summarise the news.

The dispute

Like Company runs a Hungarian news portal. It sued Google Ireland before the Budapest Regional Court, arguing that Gemini's replies to user prompts reproduced and made available content from its articles without permission. The claim relies on Article 15(1) of the 2019 Directive on Copyright in the Digital Single Market (the CDSM Directive), which lets press publishers control online uses of their publications by information society service providers. The period in dispute runs from 13 June 2023 to 31 May 2024.

The Hungarian court sent four questions to Luxembourg. Two concerned the training of the language model, and two concerned the answers the chatbot gives.

Training: not the right case

On training, Szpunar advised the Court not to answer. In his reading, the referring court had not established that the publisher's articles were actually part of the training data. He also took the view that training a model is not an "online use" of a press publication of the kind Article 15 is aimed at. Where training involves copying protected material, he said, the relevant rule is in principle the text and data mining exception in Article 4 of the same directive, which applies unless the rightholder has opted out.

So anyone hoping for a ruling on whether scraping news sites to train a model is lawful will have to wait for another case.

Answers: two conditions for reproduction

The more practical part of the opinion is about outputs. Szpunar started from a basic limit: the press publishers' right covers the way a story is expressed, not the facts or information in it. A chatbot that simply passes on what happened, in its own words, is not using the protected element.

For a reply to count as reproduction of a press publication, he suggested two things must come together. First, the provider must have obtained the content unlawfully, for example by getting around a paywall or other access restriction. Second, the reply must take up the publication in a form that a reader can recognise, to an extent that makes it unnecessary to go to the original article.

He treated the second right in play, making available to the public, slightly differently. In his view that right is engaged where the chatbot's reuse of a publication is systematically recognisable in its answers.

When the provider is liable

On responsibility, the opinion points to the provider rather than only the user who typed the prompt. A provider can be held responsible if the infringing use flows from the way its business model works, or if it has not put in place credible measures to prevent such outputs. That second limb links liability to concrete safeguards such as filters or controls on how sources are retrieved and quoted.

Why it matters beyond Europe

For publishers, the opinion offers a route to challenge chatbot answers that stand in for their journalism, but only where reuse is recognisable and substantial, and in the case of reproduction only where access was unlawful. For AI providers, it suggests that summaries in fresh wording, built on lawfully accessed pages, carry less risk, while systems that reproduce paywalled stories face more exposure.

Indonesia has no equivalent of the EU press publishers' right, and Indonesian copyright law has no specific text and data mining exception. But regional newsrooms that publish for EU audiences, and AI services offered into the EU, will be affected by the final ruling. The opinion is also likely to be cited in policy debates on AI and news in Southeast Asia, including in Indonesia's ongoing work on copyright reform.

No date has been set for the judgment.

Sources

• Court of Justice of the EU: case C-250/25, Like Company (case file)
• CEA Digital Law: Like Company (C-250/25), the Advocate General on AI chatbots and press