Hungary – Mobilguru v Google Ireland Limited, 28 July 2026

Claimant: Mobilguru Kft. (a Hungarian company publishing an online, Hungarian-language publication about mobile phones).

Defendant: Google Ireland Limited (provider of Gemini, formerly Bard, in the European Economic Area).

The Claimant publishes a Hungarian-language website providing information on mobile phones. The Defendant provides internet search and generative AI services under the name Gemini (previously Bard), including in Hungary. The Claimant had not used its website’s robots.txt file to prevent Google from accessing the publication for either search or AI purposes.

The Claimant tested Gemini with questions about mobile phones. According to the summary of the reference, Gemini identified the Claimant’s publication among recommended sources and supplied English-language information already present in its articles. When the Claimant entered URLs of its articles as prompts, Gemini produced English summaries and, when asked, identified the underlying articles as sources.

The Claimant then brought proceedings before the Fővárosi Törvényszék, arguing that these uses were not authorised. It also points to Google’s earlier use of extended news preview (ENP) licensing arrangements for press publications and states that publishers could separately prevent use for developing AI services through the robots.txt mechanism only from 28 September 2023. The Defendant disputes that its prior ENP arrangements establish liability and maintains that website operators can prevent use of their content for AI training and grounding.

Claimant’s claim. The Claimant argues that Gemini repeatedly uses its protected press content without authorisation: by reproducing it when training the chatbot and generating responses, and by making it available to the public through those responses. It relies principally on Article 15 of Directive 2019/790, Articles 2 and 3(2) of Directive 2001/29/EC and the corresponding Hungarian copyright provisions.

The Claimant refers to ENP agreements previously concluded by Google with publishers against licence fees, and to the introduction of a separate AI opt-out mechanism only in September 2023. It also argues that the CJEU’s judgment in Soulier and Doke (C-301/15) is inapplicable because that judgment preceded Directive 2019/790 and the present dispute does not concern consent without a licence fee.

Defendant’s defence. The Defendant denies any actionable reproduction during model training or the generation of responses: according to its account, the model learns linguistic patterns without retaining a store of protected content that can be recovered through prompts. It denies communication to the public on the basis that there is no “new public”, and characterises chatbot outputs as facts, individual words or very short extracts.

In the alternative, it invokes the temporary reproduction exception under Article 5(1) of Directive 2001/29/EC and the text-and-data-mining exception under Article 4 of Directive 2019/790. It argues that the former ENP arrangements were prompted by Directive 2019/790 and are not relevant to AI outputs; it further maintains that website operators can restrict training and grounding, but that Gemini does not present protected content to users.

1. Scope of an information society service. Whether that concept covers only an AI chatbot’s responses to prompts or also training and creating the underlying model when the same provider performs both activities.

2. Scope of press publishers’ rights. Whether Article 15(1) of Directive 2019/790 applies only to online uses by information society service providers and, if so, whether other uses are instead subject to authorisation by authors or other rightholders.

3. Excluded and temporary uses. If only chatbot responses qualify as the service, whether transient or incidental reproductions and uses limited to individual words or very short extracts fall outside press publishers’ powers of authorisation or prohibition.

4. Reproduction during training. If training is part of the information society service, whether press publishers’ Article 15 rights also cover potentially infringing reproductions made in the course of model training, subject to the relevant exceptions.

5. Text and data mining (TDM). Whether TDM carried out in AI training is an information society service; which acts of reproduction, extraction and retention are exempted under Article 4 of Directive 2019/790; how temporary copies and very short extracts are treated; and whether training must be distinguished from generating chatbot responses.

6. TDM and making available. Whether the rights to reproduce and extract for TDM purposes may limit the press publisher’s exclusive right to make its publications available to the public on demand.

7. Limits of the TDM exception. Whether TDM permissions are confined to reproduction, extraction and retention, or also restrict press publishers’ control over communication to those performing TDM or to chatbot users, during training and response generation; and how the temporary-copy and short-extract exclusions interact with that assessment.

8. Communication to the public by TDM users. Whether those conducting TDM may communicate press content to the public without specific authorisation, either during training or in response to prompts; how the relevant “public” should be defined; and what exceptions apply.

9. Extent of the right to opt out. Whether a publisher may exclude TDM use altogether or only uses beyond existing statutory exceptions, and how an express reservation affects reproductions otherwise covered by TDM exceptions.

10. Meaning of express reservation. Whether an express reservation is an unconditional prohibition of any use of the content and whether it prevents reliance on the TDM exception even for acts otherwise falling within Article 15 exclusions.

11. Effective means of reservation. Whether rights must be reserved in robots.txt or another machine-readable access-blocking mechanism or may instead be reserved by a clear notice on the publication or communicated to the user of the content; and whether publishers may be expected to adopt restrictions that also prevent unrelated uses.