AI and Copyright – Judicial Landscape in Germany

As generative artificial intelligence (AI) continues to transform the creation and use of copyright-protected content, German courts are increasingly confronted with the question of how established principles of copyright law should be applied to AI-related scenarios. Recent decisions now address a range of factual and legal issues, including questions relating to the copyrightability of AI output, copyright infringements arising from training datasets, models, and AI-generated output, as well as the applicability of the text and data mining (TDM) exception to dataset creation and AI training.

This Insight provides an overview of the key decisions in Germany to date, as well as upcoming court rulings that are expected to further shape the evolving legal landscape at the intersection of AI and copyright law.

KEY TAKEAWAYS

• AI-generated output must be assessed under established copyright principles with respect to both the question of protection and infringement.
• Documenting the creative process is becoming increasingly important, particularly where AI tools are involved.
• AI-generated output may be subject to copyright protection where the user exercises sufficient creative influence such that their personality is reflected in the output. The creative input can occur retroactively or gradually during the prompting.
• Modifications based on AI-generated output may qualify as a copyright-relevant adaptation only if the originality threshold is met. AI-generated output may infringe copyright where it contains essential creative elements of a protected work.
• AI models may infringe copyright according to some courts if they “memorized” a protected work (i.e., if the work used for training can be reproduced through simple prompts).
• While the TDM exception covers dataset creation, it does not extend to AI model training.

COPYRIGHTABILITY OF AI-GENERATED CONTENT

The Local Court of Munich, in a 13 February 2026 decision (Docket No. 142 C 9786/25), considered whether three AI-generated logos were subject to copyright protection. The plaintiff sought an injunction to prevent the use of AI-generated graphics, which he had created through prompts. The court held that such output could qualify as a protected work in principle and gave guidance on how copyright protection of AI-generated content is achieved. It applied the European standard according to which, for copyright protection, a work must be original and an expression of the author’s own intellectual creation.

For software-supported creations, copyright protection requires a level of human creative influence that reflects the author’s personality in the output generated. The creative input could take place retroactively or gradually during the prompting. AI-generated output could only meet the threshold of originality if (1) human prompts substantially determined the work and (2) AI merely functioned as a tool. A fully automated creation process or simply selecting an AI-generated output from among several “suggestions” is not sufficient.

According to the Munich Local Court, the logos at issue did not meet these requirements. Although the plaintiff generated them by using detailed prompts, decisive creative decisions were made by the software, not by its human user.

BURDEN OF PROOF – PARTIALLY AI-GENERATED SONG

In a 17 December 2025 decision, the Regional Court Frankfurt am Main (Docket No. 2-06 O 401/25) addressed how parties should substantiate their positions regarding copyrightability when AI involvement is alleged.

The applicant, a songwriter, sought an injunction enjoining the defendants from using certain lyrics. The songwriter claimed that she wrote the lyrics herself and later used an AI tool as part of the process of turning them into a song. The defendants argued that the lyrics were generated with AI and did not meet the requirements for copyright protection. They submitted an expert opinion, according to which the storytelling of poor craftsmanship with inconsistencies and logical gaps showed that the lyrics had also been created by AI.

The court held that if the defendants present specific, concrete indications that AI was used to create the work, the claimant must respond with a detailed account of the creative process—explaining how the work was developed and which elements are attributable to human creative input.

Based on the affidavits submitted by the claimant and the overall circumstances, the court concluded in this case that the original song lyrics were created by the claimant herself and were eligible for copyright protection as a literary work. Subsequent alterations using AI applications did not change this, even if the AI system had changed the lyrics. Any AI modifications of the texts were adaptations within the scope of protection of the original text. The court therefore found that the songwriter could assert claims for injunctive relief.

The appeal proceedings are still pending before the Higher Regional Court Frankfurt am Main.

TDM EXEMPTION APPLICABLE TO GENERATION OF DATASETS

The Higher Regional Court Hamburg, in a 10 December 2025 decision (Docket No. 5 U 104/24 – LAION v. Kneschke), considered whether the creation of an AI training dataset infringed a photographer’s copyright.

The plaintiff was a photographer whose photos were accessible on a stock photo agency’s website. The agency’s website contained an objection (opt-out) to TDM in natural language in its user and license terms and source code. The defendant, a nonprofit association, creates and offers extensive datasets for AI applications. To do so, it downloads and processes photos which are later deleted. The dataset comprises descriptions of pictures in text form and their metadata, including a link to the pictures, without including a copy of the pictures.

The court upheld the first instance judgment rendered by the Hamburg Regional Court and dismissed the plaintiff’s appeal, ruling that the reproduction was permissible under the TDM exception (Sec. 44b German Copyright Act). In this respect, the court stated that the downloading and processing of the photograph to create the dataset constituted a reproduction.

However, according to the court, this use was justified by the TDM exception, as the automated analysis of the image to extract information fell under the concept of “data mining.” It was not a requirement for the extracted information to be “hidden.” The TDM exception, which came into force in 2021 and thus prior to the advent of AI, was applicable to AI-related cases. It was expressly adopted to foster innovation, including in relation to machine learning.

The plaintiff could not prove an effective opt-out to TDM. According to the court, the natural language objection did not meet the requirement of a machine-readable format. It was not sufficient that the text can be captured by a machine, it would have been necessary that it can be interpreted by a machine in an automated process.

In addition, the court held that the wider TDM exception (without the possibility of opting out) applied because the TDM of the nonprofit organization served the purposes of research and knowledge acquisition (Sec. 60d German Copyright Act).

The further appeal is pending before the German Federal Supreme Court (Docket No. I ZR 281/25). A hearing has been scheduled for 3 September 2026.

COPYRIGHT INFRINGEMENT BY ‘MEMORIZATION’ OF WORKS IN MODEL AND REPRODUCTION IN OUTPUT

In an 11 November 2025 judgment, the Munich Regional Court (Docket No. 42 O 14139/24 – GEMA v. OpenAI) addressed whether AI training using copyrighted material without a license is permissible under German copyright law. The court also considered whether works are reproduced in the model and publicly made available when displayed as output.

GEMA, a German collecting society, brought an action against OpenAI. GEMA alleged that the large language models of OpenAI’s chatbot ChatGPT were trained on copyright-protected song lyrics without obtaining a license from the right holders and that users could reproduce the lyrics upon simple prompting.

The court found that the song lyrics at issue were “memorized” in the AI model itself, as they were reproducibly contained within the models. According to the court, this constituted an unauthorized reproduction of the works in the model. The court claimed that it was not necessary that the copyrighted work is allocated a distinct dataset within the model to meet this requirement. Rather, the determination of statistical parameters that can be used to retrieve the work with prompts was sufficient. In the opinion of the court, the coincidences between the works that served as training data and the output and the fact that the output can be triggered with simple prompts proved memorization and recreation by the model, as opposed to creation by the user.

The court rejected the application of the TDM exception with respect to model training because it considered that the exception requires that the data be subject to analysis, not storage or memorization. According to the court, the chatbot’s output also infringed upon the copyright in the lyrics because they were reproduced and made publicly available.

The appeal in this matter is pending before the Higher Regional Court Munich (Docket No. 6 U 3662/25 e).

OUTPUT RESEMBLING PHOTOGRAPHIC WORK NOT INFRINGING

The Higher Regional Court Düsseldorf, in a 2 April 2026 decision (Docket No. I-20 W 2/26), considered whether an AI-generated image that closely resembled a copyrighted photograph infringed the photographer’s rights.

In preliminary injunction proceedings, the applicant sought an injunction enjoining the reproduction of an AI-based creation of an image bearing a close resemblance to one of her photographs by the defendant. The pictures showed a dog trying to catch a ball underwater. At first instance, the Regional Court Düsseldorf held that the AI image constituted a free adaptation and rejected the applicant’s claims for copyright infringement. The Higher Regional Court Düsseldorf upheld the first instance ruling but based its judgment on different reasoning.

In its analysis, the Higher Regional Court held that the AI image at issue could not constitute a free adaptation of a photographic work. This would require that the AI image itself be a work within the meaning of copyright law. The Higher Regional Court rejected this thought, because the defendant had not exerted sufficient free and creative decisions that reflected his personality in the output. Notwithstanding this, the court rejected the claim for injunctive relief because it did not find a reproduction of the work.

In its assessment, the court applied the standard established by the Court of Justice of the European Union (CJEU) in Mio v. Konektra (judgment of 4 December 2025 – C-580/23). Instead of focusing on the overall impression of the works (which had been the German doctrine prior to the CJEU’s Mio vs. Konektra ruling), it examined whether the essential elements of the original that reflect the author’s personality have been recognizably incorporated into the AI image. In the court’s view, this was not the case, as the AI image shared only the subject of the photograph and contained neither the specific perspective nor the dynamic visual effect.

LOOKING AHEAD

It remains to be seen whether the decisions currently under appeal will withstand scrutiny by the higher courts. As generative AI is becoming increasingly embedded in all aspects of creative processes, it also appears inevitable that German courts will be faced with a growing number of AI-related copyright disputes in the near future to address the new subject matter and to further develop the applicable legal framework.

Some of the cases that will further shape the legal landscape in this area include:

• Like Company v. Google (Docket No. CJEU C-250/25): A Hungarian media company sued a search engine provider, alleging that the AI system systematically extracted and reproduced portions of copyrighted texts. The four questions referred to the CJEU by a Hungarian court include whether the operation and training of AI models infringe copyright, whether the TDM exception applies, and who should bear legal responsibility for AI-generated output. A hearing took place on 10 March 2026, and the Advocate General’s opinion is expected on 3 September 2026.
• GEMA v. Suno AI (Regional Court Munich, Docket No. 42 O 763/25): GEMA also sued Suno AI, the provider of an AI-based music generator, alleging the infringement of copyrighted work for training purposes without remunerating the right holders.
• Penguin Random House v. OpenAI: A publishing house sued OpenAI, alleging that its AI system infringes the copyright in the illustration of Der kleine Drache Kokosnuss (The little Dragon Coconut).