Case C-806/26, Manchester United Fanatics Klub – summaries, tokens and the silent publisher
A second Hungarian court has asked the Court of Justice what a chatbot owes to press publishers. Its six questions reach the parts of the problem that the first reference, Like Company, left open.
Facts
Manchester United Fanatics Klub Magyarország is suing Google Ireland Ltd before the Pécsi Törvényszék (Regional Court, Pécs) under the press publishers’ right in Article 15(1) of Directive (EU) 2019/790. The notice gives no further facts and does not explain on what basis the claimant qualifies as a press publisher. Its questions follow an LLM-based chatbot through its whole cycle: training, in which a publication is broken down into tokens and statistical relationships; the user’s prompt containing a publication; “grounding”, in which the service retrieves up-to-date material to answer questions; and the answer itself, here a newly generated summary from which the original publication “may be deduced”. The last two questions ask whether a publisher that has not expressly prohibited such use has consented to it, and whether silence means that the text and data mining exception in Article 4 applies. The case runs in parallel with Like Company (C‑250/25), also brought against Google by a Hungarian claimant.
Questions Referred
1. Must 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 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, be interpreted as meaning that making available to the public is considered to occur when a chatbot based on a large language model (LLM) operated by an information society service provider, in response to a request by the user (prompt), displays synthesised textual content, or a summary, that did not exist previously, on the basis of which the original press publication may be deduced and the content on which it is based may be recognised?
2. Must Article 15(1) of Directive 2019/790, and, in accordance with it, Article 2 of Directive 2001/29, be interpreted as meaning that the process of training an LLM-based chatbot constitutes an act of reproduction if, during that process, the content of a copyright-protected press publication is broken down into numerical tokens and statistical relationships which do not store the original content, but rather solely linguistic relationships?
3. Must Article 15(1) of Directive 2019/790, and, in accordance with it, Article 2 of Directive 2001/29, be interpreted as meaning that an act of reproduction of a copyright-protected press publication is considered to occur when the user gives an LLM-based chatbot an instruction containing a copyright-protected press publication and the chatbot processes that instruction in the context of the conversation, in order to be able to give a useful response to the user’s question?
4. If the answer to the second question is in the affirmative, does such reproduction of copyright-protected press publications fall within the scope of the exception for the purposes of text and data mining provided for in Article 4 of Directive 2019/790 when the trained LLM is subsequently used to power a chatbot capable of generating useful responses based on users’ instructions?
5. Must Article 2 and Article 3(2) of Directive 2001/29, read together with Article 15(1) of Directive 2019/790, be interpreted as meaning that the press publisher has consented to the use of a copyright-protected press publication for the purposes of the functioning of an LLM-based chatbot service – including its training and, in that context, the tokenisation of [the publication’s] content, as well as the process of grounding in order to obtain the up-to-date and factual information needed to respond to the users’ questions – when it has not expressly prohibited its use on the online platform operated by the service provider?
6. Must Article 4(3) of Directive 2019/790 be interpreted as meaning that, if a press publisher does not expressly and appropriately assert its right of exclusion, it is then considered to have authorised the use of the press publication for the purposes of the functioning of an LLM-based chatbot?
Sources
OJ notice C/2026/5481 (EUR‑Lex) · Case file on CURIA · Opinion of AG Szpunar in Like Company, C‑250/25 (ECLI:EU:C:2026:919) · Directive (EU) 2019/790 · Directive 2001/29/EC
Comment
The reference was published on the day Advocate General Szpunar delivered his Opinion in Like Company, and that Opinion answers much of it in advance. On training he proposed that the Court not rule at all. Training, he said, is not an “online use” of a press publication within Article 15, because it “occurs as part of the development of a piece of AI technology on dedicated computer infrastructure, without any information society service being provided to a user via the internet”. If the Court agrees, the second and fourth questions here fall outside the press publishers’ right for the same reason, however precisely the referring court describes tokens and statistical relationships.
On output, the Advocate General’s test turns on form of expression. A chatbot that retrieves a publication and reuses it to answer a query reproduces it only if it accessed the publication unlawfully or if “the form of expression in all or part of the press publication at issue is recognisable in the answer”, to the point that the user can dispense with the original. The first question here concerns exactly the case that test is designed to exclude: a summary “that did not exist previously” from which the original “may be deduced and the content on which it is based may be recognised”. Recognising the content is not recognising the form. On the Opinion’s approach, a summary that tells the reader what the article said, in new words, would not infringe the press publishers’ right, however much it reduced the reason to click through.
The last two questions concern silence. Article 4(3) allows rightholders to exclude text and data mining where they have “expressly reserved” their rights “in an appropriate manner, such as machine-readable means”. The sixth question asks whether a publisher that has not done so is taken to have authorised use by a chatbot. The fifth goes further and asks whether failing to prohibit use amounts to consent under the ordinary rights of reproduction and making available. That would turn an exception with an opt-out into a general rule of implied licence. The fifth question also mentions grounding, which is the stage the Advocate General treated as the relevant one for output: the service fetches a current article in order to answer.