Introduction:
On 3 April 2025, the Budapest Környéki Törvényszék (Budapest Regional Court) referred a request for a preliminary ruling to the Court of Justice of the European Union (CJEU) pursuant to Article 267 of the Treaty on the Functioning of the European Union (TFEU). The request was made in the context of proceedings between Like Company, a Hungarian press publisher, and Google Ireland Limited, a subsidiary of Alphabet Inc.
The case, Like Company v. Google Ireland Limited (Case C-250/25), centres on whether Google’s generative AI chatbot, Gemini (formerly Bard), infringes copyright by reproducing and communicating protected press content without permission.
This case could have profound implications for jurisprudence at the intersection of copyright and AI, as it is the first time the CJEU will interpret the EU’s Digital Single Market Directive’s application to generative AI technologies.
Ronin Legal analyses the case, drawing parallels with comparable U.S. lawsuits on AI and copyright.
Facts of the Case
The dispute involves Like Company, a Hungarian commercial entity that publishes and operates various news portals protected by intellectual property law.
Google Ireland Limited, part of Alphabet Inc., operates generative AI chatbot services, including Google Gemini (Bard), and the Google Search engine, available to users in the European Economic Area and Switzerland.
Gemini functions by generating new content, such as text summaries, based on patterns observed during its training, which involves processing data from sources like Google Search.
Unlike traditional databases, Gemini does not store copies of data but converts text into tokens for processing. When prompted, it can generate responses that include content from protected press publications, sometimes quoting longer passages and providing links to the original sources.
A specific instance in the case involves an article published by Like Company on its news portal, balatonkornyeke.hu, which reported on Hungarian singer Kozsó’s plan to introduce dolphins to an aquarium near Lake Balaton, alongside other details about his life, such as his hospitalization and legal issues.
When asked to summarize this article in Hungarian, Google’s Gemini provided a detailed response that included information from Like Company’s protected publication.
Like Company alleges that between June 13, 2023, and February 7, 2024, Google infringed its rights by reproducing and making its press content available to the public without consent, in violation of EU and Hungarian copyright laws.
Google, in response, sought to have the claim dismissed, arguing that its actions do not constitute infringement.
Questions Before The Court
The Budapest Környéki Törvényszék referred four key questions to the CJEU for a preliminary ruling to clarify the application of EU copyright law to generative AI:
- Communication to the Public: Does displaying text in a chatbot’s response, partially identical to a press publisher’s web content and protected under Article 15 of the DSM Directive, constitute “communication to the public” under Article 3(2) of the InfoSoc Directive? If so, does the chatbot’s predictive word-generation process affect this classification?
- Training as Reproduction: Does training an LLM-based chatbot, by observing and matching patterns to recognize linguistic structures, constitute “reproduction” under Article 2 of the InfoSoc Directive?
- Text and Data Mining Exception: If training is deemed reproduction, does it fall under the text and data mining exception provided by Article 4 of the DSM Directive?
- User Instructions and Reproduction: When a user prompts a chatbot with text matching or referencing a press publication, and the chatbot generates a response displaying part or all of that publication, does this constitute reproduction by the chatbot service provider under Article 15(1) of the DSM Directive and Article 2 of the InfoSoc Directive?
Like Company’s Stance
Like Company argues that Google’s Gemini chatbot accessed and used its protected press publications, constituting reproduction and communication to the public without consent.
The company asserts that Gemini’s responses, which includes content from its articles, exceeds the permissible use of “individual words or very short extracts” under Article 15 of the DSM Directive, causing significant economic harm by diverting users from its ad-supported news portals.
Like Company contends that this unauthorised use occurred continuously from June 13, 2023, to February 7, 2024, and that determining “very short extracts” based solely on length is inappropriate, as it could harm publishers of longer texts.
It emphasizes that Article 15 requires publisher consent for online use of press content on platforms like Google Search and Gemini, and without such consent or compensation, the use is infringing.
The company further argues that the legislative intent of Article 15 is to protect publishers’ economic interests, and allowing extensive use without compensation would undermine this purpose.
Like Company clarifies that its consent for content to appear in search engine results does not extend to Gemini’s responses, which go beyond mere referencing and constitute communication to the public, even within the chatbot’s interface.
Additionally, Like Company alleges that Google’s training of Gemini infringed its reproduction rights, as the quantity and relevance of the data, including works by Hungarian authors, exceeded the limits of any applicable exceptions.
Google’s Counter
Google argues that Gemini’s responses do not constitute reproduction or communication to the public under EU or Hungarian law, as they do not reach a “new public” within the meaning of CJEU case-law, being accessible to the same internet users who can view Like Company’s original publications.
The company asserts that the responses do not exceed the threshold of “individual words or very short extracts” permitted under Hungarian legislation. Google further contends that its training of Gemini did not occur in Hungary or use Hungarian hardware infrastructure, rendering Hungarian law inapplicable.
Even if the responses were deemed reproduction or communication to the public, Google argues they would be covered by exceptions for temporary acts of reproduction under Article 5(1) of the InfoSoc Directive and text and data mining under Article 4 of the DSM Directive, as the extracts do not exceed permissible limits.
Additionally, Google states that Gemini’s responses are not identical to Like Company’s articles, referencing only some facts and including information not derived from the protected content, often due to “hallucination,” where the chatbot, as an experimental technology, may produce inaccurate or invented content.
Google describes Gemini as a creative tool, not an information database, processing tokenized data from the Google Search database rather than storing copies, and capable of displaying modified content only if the user provides the original article in their instructions.
Finally, Google argues that Article 15 of the DSM Directive balances publishers’ rights with users’ freedom of expression and information under Article 11 of the EU Charter of Fundamental Rights and requiring service providers to monitor every extract would conflict with the principle that providers are not obliged to check web page content linked in search results.
Implications of the Case
The Like Company v. Google Ireland Limited case is the first CJEU examination of how Article 15 of the DSM Directive, which grants press publishers rights to control and seek compensation for online use of their content beyond minimal extracts, applies to generative AI like Google’s Gemini chatbot.
A ruling for Like Company could require AI developers to obtain licenses for training and deploying models in Europe, increasing costs and impacting the AI industry. A ruling for Google could restrict Article 15’s scope, allowing broader use of press content without compensation, harming publishers’ revenue.
Parallels in U.S. AI Copyright Battles
U.S. lawsuits addressing generative AI and copyright infringement share similarities with Like Company v. Google Ireland Limited, focusing on unauthorized use of protected content in AI training and outputs.
Ronin Legal covered The New York Times v. OpenAI and Microsoft (2023) in our earlier piece on AI and Ownership. The New York Times sued OpenAI and Microsoft, alleging that their AI models, including ChatGPT, were trained on its copyrighted articles without permission, infringing copyright by producing near-verbatim excerpts that compete with its business.
Unlike the EU case’s reliance on Article 15 of the DSM Directive, this case turns on U.S. fair use doctrine. As of May 2025, it remains ongoing after a federal judge denied OpenAI’s motion to dismiss in March 2025, aligning with the EU case’s focus on press content use and economic harm.
In Andersen v. Stability AI, Midjourney, and DeviantArt, visual artists Sarah Andersen, Kelly McKernan, and Karla Ortiz sued for unauthorized use of their copyrighted artworks to train image-generating AI models.
Akin to Like Company’s claim that training Gemini constituted reproduction, the plaintiffs argue this use infringes copyright.
The case examines whether training qualifies as fair use. In October 2023, the court dismissed several claims, including many direct infringements claims against Midjourney and DeviantArt (requiring more specific allegations), all Digital Millennium Copyright Act (DMCA) claims, and unjust enrichment claims.
However, crucial claims for direct copyright infringement against Stability AI (based on theories that the AI model itself constitutes an infringing copy of the training data and that distributing it is infringement) proceeded.
Claims for induced copyright infringement against Stability AI were also allowed to continue. The case is ongoing, currently in the discovery phase with a trial set for September 8, 2026, as parties continue to litigate claims related to AI training.
In our earlier piece on Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., we detailed the case and its significance for the AI industry.
In 2020, Thomson Reuters sued ROSS Intelligence for using Westlaw’s copyrighted case summaries and headnotes to train an AI legal research tool competing with Westlaw.
On February 11, 2025, a Delaware federal court, under Judge Stephanos Bibas, ruled that ROSS’s unauthorized use was copyright infringement, rejecting its fair use defence as non-transformative and market-harming.
Though not involving generative AI, this case relates to the EU case’s question on training and reproduction, suggesting U.S. courts may favour rightsholders.
