Open navigation
Search

Hungary - Like Company v Google Ireland Limited, 3 November 2026

20 Nov 2026 International 9 min read

Like Company v Google Ireland Limited

Court

Court of Justice of the European Union (CJEU)

(referring court: Budapest District Court)

CaseC-250/25, Like Company v Google Ireland Limited
CountryHungary
Parties

Claimant: Like Company Kft. (incorporated in Hungary; operator of various web portals, including https://balatonkornyeke.hu)

Defendant: Google Ireland Limited (established in Ireland; member of the Alphabet group)

Date Claim Issued

Preliminary ruling request lodged on 3 April 2025.

In the main proceedings, Claimant seeks a declaration that Defendant continuously infringed its rights between 13 June 2023 and 31 May 2024 (as described in the Opinion).

Type of ClaimInfringement of the related rights of publishers of press publications (Article 15(1) of Directive (EU) 2019/790 (CDSM Directive), transposed the Hungarian Copyright Act); unlawful reproduction and making available to the public by information society service providers
Status as of 3 November 2026Opinion of Advocate General Szpunar delivered on 3 November 2026 (provisional text). The judgment of the CJEU is pending.
Summary of Key Background Facts

Defendant provides the Gemini chatbot to users in the EU, powered by Defendant's Gemini large language models (LLMs). When asked for information, the chatbot uses Retrieval-Augmented Generation (RAG): it searches the Google Search index, uses bots to retrieve (scrape) relevant content from the webpages identified, adds it to the prompt fed into the LLM and generates an answer, sometimes with hyperlinks to the sources.

The Claimant, as a press publisher, principally published literary works of a journalistic nature which are accessible to internet users. The Claimant displays advertisements on its interfaces, for which it receives advertising revenue in proportion to the number of users who visit each interface. 

In response to a specific question, Gemini chatbot provided a detailed response, which included a summary of the contents of articles published by the Claimant.

The generative AI chatbot functions in a similar way to search engines and helps to create new content, such as images, music and code. The generative AI chatbot has been trained by means of the observation and matching of patterns (known as ‘string searching’). This makes it possible to initiate ‘conversations’ with the chatbot, in which users can obtain detailed information about, inter alia, the content of press publications, and, in certain cases, create summaries of those publications. When a chatbot directly quotes a longer passage from a web page, it highlights that page, which enables users to access the source directly with a single click.

The chatbot does not have a fixed database from which it is able to retrieve any data content at the request of users. It uses the Google Search database to collect data and usually suggests that the user search for the subject in question in Google Search afterwards. When it is asked a specific question directly, the chatbot can provide a response that displays the content of a protected press publication.

Questions Referred

In essence, the referring court asked whether

  1. the display, in an LLM-based chatbot's answers, of text partially identical to press publishers' content is a communication to the public under Article 15(1) CDSM Directive and Article 3(2) of Directive 2001/29/EC (InfoSoc Directive);
  2. training an LLM is an act of reproduction under Article 2 InfoSoc Directive;
  3. if so, the text and data mining (TDM) exception in Article 4 CDSM Directive applies; and
  4. chatbot-generated responses reproducing all or part of a press publication constitute acts of reproduction attributable to the chatbot service provider.

Written observations were submitted by the parties, nine Member States and the Commission, and a hearing was held on 10 March 2026.

Advocate General's Opinion

The Advocate General proposed that the CJEU declare the second and third questions inadmissible as hypothetical and answer the first and fourth questions together. 

Questions 2 and 3: inadmissible

According to the Advocate General, nothing supports the assumption that Claimant's articles were used for training: given the proximity between the article (21 July 2023) and the release of Gemini 1.0 (December 2023), the chatbot more likely retrieved the article in real time through RAG. Secondly, LLM training is not an "online use" under Article 15(1) CDSM Directive, which covers uses occurring as an integral part of the provision of an information society service via the internet at a recipient's request. Copyright in the underlying articles is not subject to that condition, but Claimant does not claim copyright.

Obiter: LLM training and the TDM exception

While suggesting to rule for inadmissibility, the Advocate General addressed these issues "for the sake of completeness", expressing "a provisional view only":

  • Territoriality: the Gemini LLMs were trained essentially in the United States. Szpunar considered that extending EU law to training anywhere in the world based on its effects in the EU alone would raise questions, and entertained the approach of applying EU law to scraping from servers located in the EU, and considered it preferable for the legislature to address the issue.
  • Reproduction: training involves reproductions under Article 2 InfoSoc Directive (dataset creation, pre-processing, training loop). A trained model is not a database, although outputs "regurgitating" training material may themselves be reproductions, but it should be decided on a case-by-case basis.
  • Exceptions: the temporary reproduction exception (Article 5(1) InfoSoc Directive) cannot cover the training dataset. The Advocate General states that the TDM exception does, in principle, apply to LLM training, relying on the broad definition of TDM and recital 105 and Article 53(1)(c) of the AI Act.

Three-step test: in his view, the test does not preclude that interpretation (point 113), but developers should demonstrate measures mitigating the risk of outputs closely resembling specific training material (e.g. deduplication, prompt and output filters), failing which the national court should refuse the exception. The exception is limited by lawful access (no scraping of obviously unlawful sources, no circumvention of effective technical measures) and the opt-out, which is "machine-readable" where state-of-the-art robots can understand it with sufficient reliability. Member States may provide for compensation for creators.

Questions 1 and 4: use of the chatbot through RAG

Preliminary matters: use may be presumed where the chatbot has access to the content and the claimant provides a reasonable sample of similar answers or answers linking to the source, shifting the burden to the provider. Article 15(1) applies where the chatbot is provided to users in the EU, wherever the provider or servers are located and RAG is an "online use".

Technical reproductions during RAG prima facie fall under Article 2 InfoSoc Directive (reproduction right). The temporary reproduction exception may apply if the reproductions enable a "lawful use", which depends cumulatively on (i) lawful access, without circumventing effective technical measures such as a paywall or anti-scraping measures, and (ii) the form of the answer.

The answer as a reproduction: The CJEU’s usual infringement test should apply where the cumulative criteria of copying and recognisability are examined. Copying is satisfied even though the article is fed into the prompt rather than copy-pasted. The press publishers' right protects the form of expression, not the facts, thus the decisive factor is whether the answer is a direct substitute, because the user can dispense with the original. Short summaries of the facts presented differently, or outlines drawn from several sources, are not reproductions. Making available to the public occurs only if the publication is systematically recognizable, to the point of direct substitution, in answers to related queries, which rightholders must demonstrate with sufficient examples. 

Attribution to the provider: Article 15(1) covers only use by information society service providers. Primary liability generally lies with the user who directs the chatbot, but the provider is also liable where the acts are inherent to its business model or it fails to take the appropriate measures expected of a reasonably diligent operator, such as bots respecting access restrictions and output filters.

No implied consent: leaving publications freely accessible does not amount to implied consent to RAG use. The publisher must at least be "actually and individually" informed and have simple means to prevent the use. 

Proposal of the AG 

The Advocate General proposed that the CJEU declare the second and third questions inadmissible and answer the first and fourth questions as follows:

"Article 15(1) of Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on copyright and related rights in the Digital Single Market and amending Directives 96/9/EC and 2001/29/EC, read together with Article 2 and Article 3(2) of Directive 2001/29/EC of the European Parliament and of the Council of 22 May 2001 on the harmonisation of certain aspects of copyright and related rights in the information society, must be interpreted as meaning that where, while processing a user query, an online chatbot powered by a Large Language Model (LLM) retrieves a press publication from a press publisher's website, reuses that publication to generate an answer, and provides that answer to the user, relevant acts of reproduction of the publication at issue occur if:

  • that chatbot has accessed that publication unlawfully, in particular by circumventing effective technical measures designed to restrict access to, or the scraping of, that publication; or
  • the form of expression in all or part of the press publication at issue is recognisable in the answer provided to the user by the chatbot, to such an extent that the user, by reading that answer, can dispense with accessing the original publication entirely.

An act of making available to the public of the press publication at issue occurs if the form of expression in all or part of that publication is systematically recognisable, to the extent described above, in the answers provided by the chatbot to related user queries.

Those acts of reproduction or making available may be regarded as carried out by the chatbot provider where:

  • such acts are inherent to that provider's business model; or
  • that provider has failed to put in place the appropriate measures that can be expected from a reasonably diligent operator in its situation in order credibly and effectively to counter the carrying out of such acts by way of the chatbot."
CMS Comment

AG Szpunar narrows the first CJEU AI copyright case to RAG

The Opinion proposes that the CJEU decline to rule on LLM training and the TDM exception and address only the real-time retrieval and reuse of press content by a chatbot. If the CJEU follows that approach, this case will not produce a binding ruling on training; the Advocate General's views on training are obiter.

For RAG outputs, the proposed test turns on whether the form of expression of a press publication is recognisable to the extent that the user can dispense with the original: summaries conveying the same facts in a different form would not infringe, while paraphrasing does not in itself exclude infringement. Lawful access and provider diligence are central. 

The Opinion is not binding on the CJEU, and the judgment remains to be delivered.

https://cms-lawnow.com/en/ealerts/2025/09/cjeu-to-rule-on-landmark-ai-and-copyright-case-from-hungary

previous page

Artificial Intelligence and Copyright Case Tracker

next page

2. Netherlands - DPG Media B.V.; Mediahuis Nederland B.V.; Mediahuis NRC B.V. v Knowledge Exchange B.V., trading as HowardsHome, 20 October 2026


Back to top Back to top
Warning: Fraudulent emails and messages
Opens in new window