AI in the sights: German court sets new standards for music rights 

This article has been written by Bettina Windisch, Meike Neusser-Depken, Charlotte O’Mara, and John Lane.

The German collecting society (GEMA) recently prevailed against OpenAI in court. The point of contention was the use of copyrighted song lyrics for ChatGPT training. The final decision is still pending, but the ruling is considered a signal in the discussion about copyright and AIfor the entire EU and across Europe. A second court ruling in the legal dispute between GEMA and Suno AI on music use for AI training is expected in Germany in 2026. 

GEMA vs. OpenAI: How did the Munich Regional Court decide? 

GEMA accused OpenAI of using copyrighted song lyrics without a license to train ChatGPT , including well-known German titles such as “Über den Wolken” by Reinhard Mey or “Männer” by Herbert Grönemeyer. The decisive factor here was that ChatGPT was able to reproduce almost verbatim at simple prompts. For GEMA, this indicated that the text of each work was not only used for analysis but also stored in the model (“memorised”) and thus reproduced. 

OpenAI defended itself by arguing that its models did not store any concrete data but only learned patterns and probabilities from the training data.Additionally, OpenAI invoked, among other things, the  text and data mining (“TDM”) barrier (§ 44b German Copyright Act), which allows the reproduction of copyrighted works for analysis purposes under certain conditions. 

The Munich Regional Court ruled in favor of GEMA in large part. It found that OpenAI had violated applicable copyright law with the training and operation of ChatGPT. 

The central points of the decision are: 

Reproduction by “memorisation” and “output” 

The court considered it proven that the song lyrics were reproducibly contained in OpenAI’s language models (especially GPT-4 and GPT-4o). This memorisation of the works in the model itself was assessed as a reproduction relevant to copyright within the meaning of § 16 German Copyright Act, as well as the reproduction of the song lyrics in the outputs of the chatbots. 

No coverage by TDM barriers 

In the opinion of the Chamber, the reproductions found by the court were not covered by the TDM barrier of § 44b German Copyright Act. The court argued that although this limit includes preparatory acts for analysis, it does not include permanent reproductions in the model that interferes with the exploitation rights of the authors. 

Responsibility of OpenAI 

The court emphasised that the responsibility for storing and reproducing the copyrighted content lies with OpenAI as the operator of the language models and not with the users, who had only entered simple prompts without any content specifications. 

Consequences for OpenAI 

The company was ordered to claim injunctive relief, information and damages. 

GEMA ruling: What does this mean for Austria? 

The recent GEMA decision in Germany sends a strong signal across borders, including Austria, although it is not legally binding. While there is currently no comparable lawsuit by Austria’s collecting society AKM against AI providers, similar legal actions are entirely conceivable. Due to the EU-wide harmonization of copyright law under the EU DSM Directive (2019/790/EU), Austrian courts would likely follow the German precedent if proceedings were initiated. 

AI Training and Copyright in Austria 

It’sassumed that training AI models on copyrighted song lyrics constitutes a copyright-relevant reproduction also in Austria, if the lyrics are permanently stored and can be reproduced through simple prompts. The TDM exception exists for research and cultural heritage (§ 42h Austrian Copyright Act), excluding commercial TDM. It also excludes permanent reproduction of works in outputs. Commercial TDM is prohibited if rights holders Opt-out or if outputs include substantial parts of works (§ 42h Austrian Copyright Act). Even lawful TDM does not allow full reproduction of works in outputs. Since the collecting society, AKM, has opted out for its repertoire, commercial TDM infringes copyright.  

The recent GEMA ruling will therefore strongly influence Austrian courts and practice.  AI providers storing protected texts without a license violate copyright law. The German ruling places liability on AI providers, not users generating lyrics via simple prompts, exposing companies to claims for injunctions and damages. For businesses using AI, compliance remains critical,especially Opt-out management, documentation, and auditing. 

Impact on the Music Industry in the EU 

The decision strengthens the rights of authors, composers, and publishers. Training on copyrighted material without authorization now clearly risks copyright infringement, legal liability, and financial damage across Europe. If confirmed by the German Federal Court of Justice,and possibly the ECJ,it could pave the way for new licensing models and revenue streams across the EU. However, this requires cooperation between the music industry and AI providers. A blanket Opt-out by rights holders is not a viable solution. Instead, collecting societies in Germany and Austria are expected to push for legal frameworks and licensing models that enable fair collaboration. 

Is the UK going a different way? Licensing model vs. opting-out – the discussion is open 

The Munich court’s finding that AI training on copyrighted lyrics can constitute reproduction represents the most expansive judicial interpretation of AI memorization to date in Europe. For the UK, the ruling is significant not because it has legal force here, but because it sharpens a contrast: while Germany has embraced a rights-holder-protective view that equates memorization with reproduction and thus copying, the UK High Court has taken a more cautious, evidence-driven approach, as seen in Getty Images v Stability AI  – highlighting scepticism that AI model weights constitute “copies” absent clear, extractable reproduction. 

The German decision has important signalling effects for UK policy and industry because it shows how a major EU Member State interprets the EU’s TDM rules in a rights-holder-centric way,and that interpretation contrasts with the UK’s currently ambiguous position. 

Under the EU’s DSM Directive, commercial TDM is permitted subject to a rights-holder Opt-out. Germany’s judgment effectively reinforces the rights of creatives in the AI arena, signalling that rights-holders who exclude their works from TDM can expect courts to take that exclusion seriously,and potentially treat training on those works without a licence as infringement. 

By contrast, the UK’s TDM rules (S29A Copyright Designs & Patents Act 1988) are under discussion and less tested. While the UK government previously flirted with an expansive TDM exception for commercial AI, this was later withdrawn under pressure from creative industries – requiring clearer transparency in relation to AI training, rights-holder control over use of works and fair remuneration and strong calls for a fair licensing system. The UK finds itself in a transitional state without clear opt-in or Opt-out mechanisms for large-scale AI model training. 

Unlike the EU the UK has no commercial TDM exception, no statutory Opt-out and no clear licensing framework for training data. The GEMA decision shines a light on the UK’s indecision.  Whilst the UK creative industries have largely argued against an Opt-out mechanism as it would create too much of a burden on creators and rights-holders to be workable, this judicial decision could be seen to bolster arguments for enforceable rights on training whichever system is adopted, provided there is clarity. 

Practically, the German ruling may not change UK doctrine in the short term, but it could be seen to strengthen the lobbying position of UK rights-holders who argue that the existing status quo undervalues the creative sector’s interests in the training stage. It raises expectations that the UK will need to define a more coherent position on AI training and the future balance between innovation and copyright protection—either through legislation or future case law,if it wishes to avoid a widening regulatory gap with major European markets.  

In the USA, the music industry has leaned toward making business deals instead of litigation – favoring a licensing model 

At the present time in the US, there are dozens of pending cases brought by rights holders over whether it is fair to use copyrighted material without payment for AI training, and many attorneys still expect that the U.S. Supreme Court may have to weigh in at some point to finally settle the issue. 

There have only been a few US district courts that have ruled in the context of summary judgment proceedings (as contrasted with judgments after a full trial) and no regional circuit court of appeals has yet ruled on whether such uses amount to a “fair use” or copyright infringement under US law.  As the doctrine of “fair use” under US law is a very fact-specific inquiry, it is dangerous to extrapolate the result in one case for other cases.  

Turning conflict into collaboration 

In the context of music-related copyright however, the trend thus far in the US has weighed on the side of resolving such conflicts through settlement and ongoing collaborative arrangements.  For example, Universal Music Group (UMG) and Stability AI have announced recently a strategic alliance to co-develop professional, licensed AI music creation tools.  

The agreement focuses on building “next-generation professional music creation tools, powered by responsibly trained generative AI” that support the creative process of artists, producers, and songwriters. Both companies have emphasized an artist-centric approach, ensuring the tools are fully licensed and commercially safe, prioritizing input from the creative community.  

This partnership highlights a broader shift in UMG’s strategy, which has historically involved legal challenges against AI companies that use copyrighted material without authorization for training their models.  

For example, in addition to the Stability AI settlement, UMG has officially resolved its copyright dispute with AI music generator Udio, ending one of the most closely watched legal battles in the music industry.   The dispute was resolved by way of a separate licensing agreement in lieu of further litigation, revealing that they will now collaborate on an AI-driven music creation and streaming platform scheduled for release in 2026. 

The UMG/ Udio lawsuit, filed more than a year ago, accused Udio and fellow AI developer Suno of using millions of copyrighted songs from major record labels without authorization to train their systems. While Sony Music Group and Warner Music Group continue their cases against Udio, UMG’s decision to settle shifts the focus from litigation to innovation. 

According to both companies, the forthcoming platform will use UMG’s licensed catalog to train its models, allowing users to generate and customize music in a fully authorized environment.  

Another high profile settlement of a class action literary copyright  dispute  

In the high-profile case Bartz et al. v. Anthropic PBC, a U.S. District Court ruled that while Anthropic’s use of copyrighted material to train its AI model Claude constituted fair use, the company was liable for using pirated books. Anthropic subsequently agreed to a preliminarily approved settlement of $1.5 billion, equating to approximately $3,000 per copyrighted work by the class members. The settlement also mandates the destruction of pirated book datasets and constitutes a “past release only,” meaning it covers claims for conduct predating the settlement, with limitations regarding future reproductions, distributions, derivative works, and product outputs. 

Conclusion:  

The Munich Regional Court has set a European precedent by ruling that AI “memorization” and output of copyrighted works constitute unlawful reproduction, influencing Germany, Austria and the wider European Union. In contrast, the United Kingdom needs to set a clear position on AI training and find a fair balance between supporting innovation and protecting copyright either through new legal framework or court decisions to avoid a gap with major European markets. Meanwhile, in the USA, where the Fair Use Doctrine creates uncertainty, the music industry has increasingly favored licensing agreements and collaborations to resolve conflicts. 

Whether licensing model or Opt-out, these diverging approaches highlight that the consequences for the music industry are far-reaching: AI providers must ensure their systems have been legitimately trained, the music industry must actively exercise its rights and secure licensing revenues, and users should also be informed of the complex and to a significant extent, unsettled legal implications. Given this complex and evolving legal landscape, seeking legal advice is essential to protect copyrights and mitigate risks arising from AI use. 

If you need expert legal advice, contact Bettina Windisch Alteri here.