{"id":1789,"date":"2026-06-15T09:55:46","date_gmt":"2026-06-15T09:55:46","guid":{"rendered":"https:\/\/t-s.legal\/news\/the-historic-ruling-of-the-regional-court-in-munich-in-the-gema-vs-openai-case-on-copyright-and-training-artificial-intelligence\/"},"modified":"2026-06-15T12:57:56","modified_gmt":"2026-06-15T12:57:56","slug":"the-historic-ruling-of-the-regional-court-in-munich-in-the-gema-vs-openai-case-on-copyright-and-training-artificial-intelligence","status":"publish","type":"post","link":"https:\/\/t-s.legal\/en\/views\/the-historic-ruling-of-the-regional-court-in-munich-in-the-gema-vs-openai-case-on-copyright-and-training-artificial-intelligence\/","title":{"rendered":"The Historic Ruling of the Regional Court in Munich in the GEMA vs. OpenAI Case on Copyright and Training Artificial Intelligence"},"content":{"rendered":"\n

The Historic Ruling of \nthe Regional Court in \nMunich in the GEMA \nvs. OpenAI Case on \nCopyright and Training \nArtificial Intelligence<\/strong><\/p>\n\n

The German collective management organization for \nmusic authors\u2019 copyright (GEMA) filed a lawsuit with \nthe Regional Court in Munich against OpenAI, accusing \nthe company of using protected lyrics from nine popular \nGerman songs without authorization in the training of \nits large language models, GPT-4 and GPT-4o. Among \nthose songs are \u201cAtemlos\u201d by Kristina Bach, \u201cM\u00e4nner\u201d \nby Herbert Gr\u00f6nemeyer, and \u201c\u00dcber den Wolken\u201d by \nReinhard Mey.<\/p>\n\n

GEMA argued that these lyrics are stored in the model\u2019s \nparameters and can be reproduced almost identically, \nwhich would constitute unauthorized recording and reproduction under the provisions of the German \nCopyright Law. <\/p>\n\n

OpenAI, on the other hand, maintained \nthat these models do not store specific texts or data, \nbut instead reflect statistical patterns learned during \nthe analysis of the entire dataset. According to their \ninterpretation, the content the models generate is the \nresult of user prompts, and control over that content does \nnot rest with them. <\/p>\n\n

OpenAI also stated that its practices \nfall under the exceptions provided by the Directive on \nCopyright and Related Rights in the Digital Single Market \n(EU) 2019\/790 (the CDSM Directive), which regulates \nexceptions for text and data mining, and that these \nexceptions cover the training of AI models.<\/p>\n\n

What did the Munich Regional Court decide?<\/p>\n\n

The first-instance ruling in this case is of great importance, \nas it largely upheld GEMA\u2019s claims, including prohibiting \nfurther reproduction and public communication of the \ncontent, as well as awarding damages. <\/p>\n\n

The court stated \nthat simple user prompts can lead ChatGPT to reproduce \nlarge portions of the original texts almost identically. \nAlthough some \u201challucinations\u201d were observed in certain \nresponses, the court held that this does not diminish the \nidentifiability of the original texts, since the memorized \ncontent had not changed in any substantial way, and \nthe differences were mostly limited to introductory or \nconcluding parts of the texts.<\/p>\n\n

The scope and complexity \nof the generated content demonstrated that this was not \na matter of coincidence. In fact, the parties agreed in \nthis proceeding that the song lyrics were used in training \nthe models, but they disagreed on whether, from a \nlegal standpoint, this constituted authorized recording, \nreproduction, and public communication of the works. <\/p>\n\n

The Storage of Content as a Form of Copyrighted Work Reproduction<\/strong><\/p>\n\n

The court relied on scientific studies in the field of \ninformation technology, which indicate that training \ndata can exist within a model\u2019s parameters and remain \naccessible \u2013 a phenomenon that GEMA referred to as \nmemorization.<\/p>\n\n

According to the court\u2019s findings, if \ncontent can be fixed in a mathematical form, whether through numerical probability values or by some other \ntechnical means, such fixation (recording) can be \nconsidered reproduction, that is, copying of the work. \n<\/p>\n\n

The \ncourt further established that simple user requests \u2013 such \nas \u201cWhat are the lyrics to the song [title]?\u201d or \u201cWhat is the \nchorus of the song [title]?\u201d \u2013 can lead to reproduction of \nthe content, and this fact was decisive in the conclusion \nthat such activities by OpenAI also constitute the \nrecording of a copyrighted work.<\/p>\n\n

Rejecting OpenAI\u2019s \nclaim that GEMA must identify the specific parts of \nthe text stored within the model, the court emphasized \nthat it is sufficient for the model to be able to generate \nstatistically likely sequences that recognizably reproduce \nthe song lyrics based on patterns learned during training<\/p>\n\n

As mentioned earlier, based on these findings, the \ncourt concluded that memorizing song lyrics within the \nparameters of an artificial intelligence model is equivalent \nto the recording of a work, and that reproducing such \ncontent through ChatGPT constitutes the acts of \nreproduction and public communication.<\/p>\n\n

Given that \nOpenAI did not obtain authorization from the rights \nholders, its activities in the process of training AI \nmodels, as well as the subsequent use of those models by \nusers, constitute unauthorized reproduction and public \ncommunication of copyrighted works.<\/p>\n\n

Are text and data mining exceptions applicable to AI model training?<\/strong><\/p>\n\n

As we mentioned at the beginning, OpenAI also based \nits defense on the exceptions provided by the CDSM \nDirective, claiming that training AI models is covered \nby the copyright exception for text and data mining.<\/p>\n\n

In practice, in the absence of other regulation, the rule \nsuspending copyright for the purpose of text and data \nmining (TDM) is often applied to the training of artificial \nintelligence. <\/p>\n\n

TDM is the process of automatically or semiautomatically analyzing large volumes of texts or data in \norder to discover patterns, information, or knowledge \nthat are not immediately apparent and that can provide \nuseful insights for scientific and other forms of research. <\/p>\n\n

Thus, Article 3 of the Directive provides that research \norganizations and institutions such as universities and \nmuseums may, for scientific purposes and without the \npermission of rights holders, perform text and data \nmining. Article 4 extends this possibility to commercial \ntext and data mining, provided that the content has been \nlawfully obtained and that the rights holders have not explicitly prohibited such use, for example, through \nmachine-readable terms.<\/p>\n\n

In short, these articles allow for \nscientific, research, and even commercial text and data \nmining, subject to certain conditions and with respect for \nthe rights of content owners. <\/p>\n\n

Why did the court reject the reliance on the TDM exceptions?<\/strong><\/p>\n\n

However, the court concluded that the suspension of \nrights for text and data mining cannot be applied in this \ncase, since the training of large language models does \nnot consist solely of data analysis, but also of their direct \nreproduction. <\/p>\n\n

Exceptions in the legislation are intended \nfor processes of research and information analysis, not \nfor recording and reproducing specific protected works, \nwhich is the case here. <\/p>\n\n

The memory of an artificial \nintelligence system, which enables the reproduction of \ncopyrighted works through simple prompts, exceeds the \nscope of the purpose these exceptions are meant to serve, \nand therefore, the court considers that their application \nis not justified. <\/p>\n\n

OpenAI’s Liability and Possible Measures to Prevent Copyright Infringement<\/strong><\/p>\n\n

It was also emphasized that responsibility for such \nactivities cannot fall on the users of the model, but \nrather on the development teams and the companies that \ndevelop these models. <\/p>\n\n

In addition, it was emphasized that \nit is difficult to remove specific data from already trained \nmodels, but that it is nevertheless necessary to establish \nmeasures to prevent future violations, such as internal \nguidelines, filters, additional licenses, or retraining of the \nmodels.<\/p>\n\n

The Significance of the Judgment for the Future of Artificial Intelligence and Copyright Law<\/strong><\/p>\n\n

Based on these legal interpretations, the ruling is \nexpected to influence the legal framework governing \nthe development and use of artificial intelligence in \nthe future, particularly concerning the use of protected \ncontent without permission. <\/p>\n\n

Although the situation is still \nuncertain and far from final, as OpenAI has announced \nthat it will appeal, GEMA is at the same time conducting \nanother lawsuit against the company Suno AI, which \nconcerns music generated by artificial intelligence.<\/p>\n","protected":false},"excerpt":{"rendered":"

The Historic Ruling of the Regional Court in Munich in the GEMA vs. OpenAI Case on Copyright and Training Artificial Intelligence The German collective management organization for music authors\u2019 copyright (GEMA) filed a lawsuit with the Regional Court in Munich against OpenAI, accusing the company of using protected lyrics from nine popular German songs without […]<\/p>\n","protected":false},"author":3,"featured_media":1788,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"slim_seo":{"title":"The Historic Ruling of the Regional Court in Munich in the GEMA vs. OpenAI Case on Copyright and Training Artificial Intelligence {{GEMA vs. OpenAI. Is training AI models a copyright infringement? sep }} - T-S Legal","description":"The Historic Ruling of the Regional Court in Munich in the GEMA vs. OpenAI Case on Copyright and Training Artificial Intelligence The GThe Regional Court in Munich issued a significant ruling on the use of copyrighted texts for training AI models. Analysis of the judgment and its implications."},"footnotes":""},"categories":[18],"tags":[],"class_list":["post-1789","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-views"],"_links":{"self":[{"href":"https:\/\/t-s.legal\/en\/wp-json\/wp\/v2\/posts\/1789","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/t-s.legal\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/t-s.legal\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/t-s.legal\/en\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/t-s.legal\/en\/wp-json\/wp\/v2\/comments?post=1789"}],"version-history":[{"count":0,"href":"https:\/\/t-s.legal\/en\/wp-json\/wp\/v2\/posts\/1789\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/t-s.legal\/en\/wp-json\/wp\/v2\/media\/1788"}],"wp:attachment":[{"href":"https:\/\/t-s.legal\/en\/wp-json\/wp\/v2\/media?parent=1789"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/t-s.legal\/en\/wp-json\/wp\/v2\/categories?post=1789"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/t-s.legal\/en\/wp-json\/wp\/v2\/tags?post=1789"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}