Introduction
The EU has followed the U.S. lead, albeit with a much shallower version of the exception as far as businesses are concerned. The Directive on Copyright in the Digital Single Market (DSM Directive), adopted in 2019, introduced two mandatory exceptions under EU copyright law. The first is for research and cultural organizations to conduct research (Research Exception), and the second is available to any type of beneficiary for any manner of use, but with a significant caveat – it may be overridden by “opt-out,” a concession to rightsholders introduced during the very last stage of the DSM Directive’s adoption process (General Exception).
EU text and data mining (TDM) law is moving from theory to practice: Courts are testing how the TDM exceptions apply to AI training and chatbot outputs, while the European Commission is looking for clearer technical rules for copyright opt-outs under the General Exception.
Text and data mining in the EU: A tale of two exceptions
Article 2(2) of the DSM Directive1 defines TDM as “any automated analytical technique aimed at analyzing text and data in digital form in order to generate information which includes but is not limited to patterns, trends and correlations.”2
The DSM Directive created two TDM exceptions:
-
Research Exception: available for research organizations and cultural heritage institutions;3 and
-
General Exception: available for everyone else (General Exception).4
The Research Exception is narrower in who can use it, but stronger once it applies. The broader General Exception can support commercial TDM uses, including some AI-related uses, but only where the rightsholder has not reserved its rights by declaring an “opt-out.” Neither of the two TDM exceptions is available where the content was unlawfully accessed.
The opt-out question is becoming practical
The broad General Exception is the provision most relevant to commercial AI training. According to the DSM Directive, for online content, rights reservations (opt-out) must be made by machine-readable means.5
Furthermore, the EU AI Act6 adds pressure: providers of general-purpose AI models must have policies to comply with EU copyright law, in particular to identify and comply with, including through state-of-the-art technologies, opt-outs from the General Exception declared by rightsholders.7
In December 2025, the European Commission consulted on protocols for reserving rights from TDM, with the aim of identifying solutions that are technically workable and widely adopted.8 The European Commission is also reviewing the DSM Directive and has launched a targeted initiative for a better copyright environment for European creativity and innovation.9 That process does not directly change current law, but it shows that EU policymakers see AI, copyright enforcement, and innovation as connected issues.
Courts are starting to draw the lines
The most important EU case is a reference for a preliminary ruling currently pending before the Court of Justice of the European Union (CJEU) in case C-250/25 (Like Company).10 A Hungarian publisher claims that an LLM-based chatbot generated summaries using protected news content. The CJEU has been asked whether chatbot outputs can infringe copyright or press publishers’ rights, whether LLM training constitutes a copyright-relevant act of reproduction, and, if so, whether the General Exception can apply. The CJEU held an oral hearing in March 2026, and its judgment is likely to matter well beyond the facts of the case.
German courts point in different directions: According to the Hamburg Court of Appeal (Hanseatisches Oberlandesgericht Hamburg), creating an AI training dataset may fall within both the Research Exception and the General Exception, while a natural-language opt-out would not be sufficient for online content; in the view of the court, such natural-language opt-out would not necessarily be machine-readable.11
By contrast, in Munich, the District Court I (Landgericht München I) reached a more rightsholder-friendly result in a case brought by the German collecting society GEMA against a provider of an LLM. The case concerned song lyrics that the court found had been memorized in the LLM and reproduced in outputs. The court held that this could amount to copyright infringement and that the General Exception would not excuse the permanent reproduction of protected lyrics in the model.12
In a similar case brought by GEMA against a provider of an AI music generator, Landgericht München I again ruled in favor of rightsholders.13 The case concerned musical works that the court found had been memorized in the LLM and reproduced in the output. The court held that compliance with the requirements under the EU AI Act does not relieve providers of their obligations under copyright law. In the court’s view, the EU AI Act does not derogate from EU copyright law; rather, it complements it. The court also considered the General Exception and held that it was not available: According to the court, the General Exception would only permit reproductions made prior to the training process.
Both decisions of Landgericht München I are first instance, but they serve as a reminder that output risk and memorization risk are not just theoretical.
Outlook
The EU position is more structured than the U.S. fair use debate, but also more uncertain until the CJEU and national courts provide clearer answers.
In the meantime, the practical lesson for AI providers is to keep evidence of lawful access, monitor machine-readable reservations, and reduce the risk that models reproduce protected works. For rightsholders, broad website terms may not be enough: opt-outs should be expressed in a format that automated systems can identify and follow.
1. Available at eur-lex.europa.eu.2. Article 2(2) DSM Directive.3. Article 3 DSM Directive.4. Article 4 DSM Directive.5. Recital (18) and Article 4(3) DSM Directive.6. Regulation (EU) 2024/1689 of the European Parliament and of the Council of June 13, 2024, laying down harmonized rules on artificial intelligence and amending Regulations (EC) No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139 and (EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (Artificial Intelligence Act), available at eur-lex.europa.eu.7. Article 53(1)(c) EU AI Act. In addition, providers of general-purpose AI models must draw up and make publicly available a sufficiently detailed summary about the content used for training (Article 53(1)(d) EU AI Act).
8. European Commission, “Commission launches consultation on protocols for reserving rights from text and data mining under the AI Act and the GPAI Code of Practice,” December 1, 2025, digital-strategy.ec.europa.eu.
9. European Commission, “Commission seeks views on the review of EU copyright rules,” May 18, 2026, digital-strategy.ec.europa.eu; European Commission, “Targeted initiative for a better copyright environment for European creativity and innovation,” ec.europa.eu.10. infocuria.curia.europa.eu/.
11 Hanseatisches Oberlandesgericht Hamburg, judgment of December 10, 2025, 5 U 104/24, landesrecht-hamburg.de (German content).12. Landgericht München I, judgment of November 11, 2025, 42 O 14139/24, gesetze-bayern.de (German content); see also Reed Smith Viewpoints, Copyright & AI: German court issues landmark decision in favor of rights holders, November 11, 2025, reedsmith.com.13. Landgericht München I, judgment of July 31, 2026, 42 O 763/25, gesetze-bayern.de (German content), see also Reed Smith Viewpoints, GEMA notches a second transatlantic AI copyright win in Germany, reedsmith.com.