Munich Court delivers landmark ruling against OpenAI

The Munich Regional Court has handed down a landmark judgment (case no. 42 O 14139/24) holding OpenAI directly liable for copyright infringement in both training and output of its GPT-4 and GPT-4o models. In GEMA (German collective society for performance and reproduction right) v OpenAI, the court found that ChatGPT reproduced verbatim lyrics from nine well-known songs by German artists including Herbert Grönemeyer and Helene Fischer. According to the judges, this was not the result of abstract pattern learning but of the models memorising and storing protected texts, constituting unlawful reproduction under German copyright law.

In this regard two independent infringements occurred: the storage of copyrighted lyrics during model training, and the unauthorised generation of those lyrics upon user request. OpenAI’s reliance on the EU text-and-data-mining exception was rejected since the exception does not permit retention of works in a form that enables full reconstruction. GEMA consequently secured injunctive relief against OpenAI, including further annex claims. However, their claims relating to authors’ personality rights were dismissed, as the court found the factual and legal threshold for those claims unmet.

The judgment is of fundamental importance; however, the German Supreme Court may have the last word. The appeal is already pending with the Appellate Court. Notably, the Munich Regional Court declined to refer the case to the Court of Justice of the European Union (CJEU), signalling confidence that current EU copyright rules provide sufficient clarity.

Stability AI (mostly) prevails in UK High Court claim

By contrast, in the UK, the High Court recently rejected Getty Images’ secondary copyright infringement claim against Stability AI, only making limited trade mark infringement findings in Getty’s favour.

Getty had initially alleged that the scraping and use of its copyright-protected images during the development and training of the generative AI model Stable Diffusion constituted primary copyright infringement. However, by trial, the primary copyright infringement claims had been abandoned by Getty, as it transpired that the relevant acts, including training and development, occurred outside the UK.

Addressing the secondary copyright infringement claim, the court found that an “infringing copy” under sections 22, 23, and 27 of the Copyright, Designs and Patents Act 1988 could contain intangible articles. However, it rejected Getty’s argument that Stable Diffusion itself was an infringing copy, because the model weights (the learnable parameters controlling the functionality of the AI model) within Stable Diffusion did not themselves store or reproduce any of Getty’s copyright-protected images.

Getty was slightly more successful in its trade mark infringement claims under sections 10(1) and 10(2) of the Trade Marks Act 1994. The court found that, in a limited number of cases, Stable Diffusion produced outputs that were identical or similar to the GETTY IMAGES or ISTOCK trade marks, and there was a likelihood of (post-sale) confusion. However, these findings were confined to specific historical versions of Stable Diffusion.

Getty’s section 10(3) trade mark infringement failed, among other things, because there was no evidence of a change in consumers’ economic behaviour. Indeed, the court noted that consumers searching for free, non-watermarked Getty images would not find Stability AI’s images bearing GETTY or ISTOCK watermarks to be of interest.

Key takeaways

Whilst UK legislation is “always speaking” (the principle that, as a general rule, a statute should be interpreted taking into account changes that have occurred since the statute was enacted), the High Court could not extend the concept of an “infringing copy” to fill perceived policy gaps. In the meantime, the judgment serves as a reminder that infringement findings are territorial, evidence-driven and fact-specific.

Getty Images has now been granted permission to appeal the UK High Court decision with the trial judge noting that it raises a “novel and important” point of copyright law. It will be very interesting to see how the Court of Appeal grapples with the statutory meaning of “infringing copy”.

While the High Court deferred the broader policy questions to Parliament, the Munich ruling reinforces an emerging EU stance that strengthens copyright as a property right. Yet for now it remains a first, not definitive step in EU AI jurisprudence. The Munich ruling underscores that large-scale AI models are not exempt from copyright constraints and highlights the growing expectation that developers implement robust licensing frameworks for creative content.

Case details at a glance

Jurisdiction: Germany
Decision level: Munich Regional Court
Parties: GEMA v OpenAI
Date: 11 November 2025
Citation: 42 O 14139/24
Decision: dycip.com/LGMunich-O-14139-24

Jurisdiction: England & Wales
Decision level: High Court
Parties: Getty Images (US) Inc & Ors v Stability Ai Limited
Date: 04 November 2025
Citation: IL-2023-000007
Decision: dycip.com/2025-EWHC-2863-Ch