Fraud Blocker AI chatbots and press publishers' rights - Hortis Legal

AI chatbot reusing news articles? That may infringe press publishers' rights

By Rachaad Farzand Ali ·

In Like Company v Google (C-250/25), Advocate General Szpunar gives a substantive answer for the first time: training a language model is permitted in principle, but a chatbot that retrieves press articles in real time and reproduces their wording may infringe press publishers' rights.

AI chatbot reusing news articles? That may infringe press publishers' rights

May an AI chatbot use news articles to answer users' questions? Advocate General Szpunar of the Court of Justice of the EU has given a substantive answer to that question for the first time. In his Opinion of 3 November 2026 in Like Company v Google (C-250/25), he draws a clear distinction:

  • training a language model is permitted in principle;
  • but a chatbot that retrieves press articles in real time and reproduces their wording may infringe.

What are press publishers' rights?

Since the 2019 DSM Directive, publishers of press publications have had a right of their own (Article 15). It allows them to take action against the online use of their publications by information society service providers, such as news aggregators and search engines. In the Netherlands, this right has been incorporated into the Neighbouring Rights Act (Wet op de naburige rechten).

The right expires two years after publication. Hyperlinks, very short extracts and private use are not covered. Until now, the question was mainly how the right relates to search engines. Generative AI has added a new question.

What is the case about?

Hungarian company Like Company operates news websites. It claims that Google's chatbot Gemini used its articles to train the underlying language models, and that Gemini reproduced those articles in its answers and made them available.

Gemini uses so-called Retrieval-Augmented Generation (RAG): when a question is asked, the chatbot retrieves up-to-date information from the internet and incorporates it into its answer. The Budapest court referred four questions to the Court of Justice: two on training and two on the chatbot's answers.

Training: permitted in principle

The AG considers that the Court need not answer the questions on training. There is no evidence that Like Company's articles were used for training, and in his view training is not a form of online use of press publications. The questions are therefore hypothetical in this case.

For the sake of completeness, he notes that training a language model on protected works is in principle covered by the text and data mining (TDM) exception in Article 4 of the DSM Directive. Note: that exception does not apply if the rights holder has reserved such use in an appropriate manner, for example in a machine-readable way.

Real-time answers: this is where it can go wrong

The position is different for the answers themselves. If a chatbot retrieves a press publication, uses it to generate an answer and provides that answer to the user, press publishers' rights may be infringed in certain circumstances.

The AG stresses that mere facts and news data are not protected. The right protects the publisher's investment and the specific form of expression of the article, not the news itself.

What should the court look at?

To determine whether there is infringement, the AG identifies two factors:

  1. Lawful access: did the chatbot retrieve the article by circumventing effective technological measures, such as a paywall or an anti-scraping block?
  2. Nature of the answer: does the answer merely convey facts, or does it recognisably reproduce the article's form of expression, to the point that the user no longer needs to consult the original article?

There is in principle a communication to the public if the form of expression systematically and recognisably reappears in answers to similar user questions.

When is the provider liable?

The chatbot provider is liable if these acts are inherent in its business model. It is also liable if it has not taken appropriate measures to counter them credibly and effectively. Arguing that 'the AI did it on its own' therefore offers little protection.

Is this the final word?

No. An Advocate General's Opinion is advisory and does not bind the Court. The judges have only just begun their deliberations; the judgment will follow later. The Court often follows the AG, but not always. Even so, this is the first substantive EU roadmap for generative AI and press publishers' rights, and parties will already be taking it into account.

What does this mean for you?

Are you a publisher or do you publish content online?

  • Record your text and data mining opt-out in a machine-readable way, for example in robots.txt and your terms of use.
  • Document the technological measures you use, such as a paywall or anti-scraper blocks.
  • Regularly test whether AI chatbots recognisably reproduce your wording or make a visit to your article unnecessary.
  • Keep evidence of this, such as dated screenshots and the questions asked.

Do you develop or use AI applications?

  • In AI audits, distinguish between training, real-time retrieval and the generated output.
  • Respect opt-outs, paywalls and scraping restrictions, and record that you do so.
  • Take measures against verbatim or recognisable reproduction of source texts.
  • In AI contracts, include separate warranties, logging, a notice-and-action procedure and indemnities for both training data and RAG output.

Need help?

Hortis Legal advises on copyright, neighbouring rights and the use of AI within your business. Would you like to know whether your content is being reused by AI applications, or would you like your own AI use and contracts reviewed? Feel free to contact us.

Source: Opinion of Advocate General Szpunar of 3 November 2026, Case C-250/25, Like Company, and Court of Justice press release No 155/26.

Questions?

Feel free to contact us

More news