OpenAI faces legal setback as German court rules it infringed on song rights

Depositphotos

The Munich court delivers a landmark ruling

A Munich regional court found that ChatGPT violated German copyright law by reproducing protected song lyrics, concluding that OpenAI trained on nine German hits without authorization.

The decision orders undisclosed damages and establishes that memorizing and outputting lyrics can infringe exploitation rights.

It is among the clearest European rulings to date that both training ingestion and output reproduction can trigger copyright liability, putting licensing obligations squarely on the table for AI developers using protected creative works.

Sound engineer working on computer producing music

Depositphotos

How the case began with GEMA’s push

Germany’s music rights society, GEMA, filed the suit in November 2024, representing lyricists, composers, and publishers with around 100,000 members.

They argued ChatGPT could reproduce lyrics because models had learned them during training. The court agreed that use requires permission and compensation, thereby strengthening the hands of collecting societies in negotiations with AI firms.

It’s a procedural path that we may see replicated across European jurisdictions as creators test how far training exemptions can be stretched in the era of AI.

hurricanehank/Depositphotos

Nine songs that shaped the dispute

The ruling references nine well-known German songs, including Herbert Grönemeyer’s Männer. Some reports also mention other hits such as Helene Fischer’s Atemlos Durch die Nacht.

By naming specific works, the court underscored that both the ingestion and subsequent reproduction of those lyrics mattered.

Factual anchoring helps future claims: plaintiffs can point to individual tracks and demonstrate availability in outputs. It also signals to AI companies that “generalized learning” defenses may falter when concrete examples of memorized text are shown.

Depositphotos

The judge emphasizes exploitation rights

Presiding judge Elke Schwager determined OpenAI infringed copyright exploitation rights, finding that storing protected lyrics in model parameters and reproducing them in responses breaches the law.

The judgment rejects the idea that only user prompts create liability, placing primary responsibility on the model operator.

That stance could ripple across cases involving images, code, and news text, where the risks of memorization and regurgitation are similar. It raises the bar for technical safeguards against verbatim outputs.

Woman using a mobile phone with ChatGPT on the screen.

AlexRuhl/Depositphotos

OpenAI’s core defense and why it fell short

OpenAI argued that ChatGPT doesn’t store training data verbatim and that outputs are generated from statistical patterns, making users responsible for any infringing prompts.

The court disagreed, stating that memorization and reproduction within the model are attributable to the provider.

This narrows the “user liability” escape hatch and shifts compliance burdens upstream to model builders, whose datasets, filters, and output controls now face legal scrutiny in Europe.

Judge gavel and law books in court law and justice

Depositphotos

Damages and remedies are still developing

While the court ordered damages, it didn’t disclose a figure. Legal observers expect a licensing conversation to follow, with GEMA hopeful about frameworks that compensate rights holders when models learn from and regenerate lyrics.

Injunctive relief and information rights were reportedly upheld, suggesting that OpenAI may need to disclose certain aspects of its training and usage to enable accurate royalty calculations and accountability, a lever that creators have sought for years.

Shutterstock

Why Grönemeyer’s songs mattered

Anchoring the case in beloved cultural touchstones like Grönemeyer’s Männer and Bochum made the legal issue instantly relatable.

It demonstrated how a chatbot’s impressive recall can cross a bright line when it surfaces verbatim lyrics.

For the court, these examples weren’t edge cases; they were proof that training had ingested and could reproduce copyrighted text. That narrative clarity can influence public sentiment and policymaking beyond the courtroom.

Depositphotos

Law firm perspectives shape the narrative

Raue, counsel to GEMA, called the judgment a first-of-its-kind confirmation in Germany that AI providers can’t use lyrics to train if outputs reproduce them absent a license.

Commentary from IP firms suggests the 42nd Civil Chamber essentially granted GEMA’s claims for injunctions, information, and damages.

These practitioner insights hint at how remedies may be enforced and how similar claims could be structured elsewhere in the EU.

OpenAI headquarters glass building in San Francisco, USA

Skorzewiak/Depositphotos

A likely appeal but limited immediate scope

OpenAI stated that it disagrees and is considering its next steps, emphasizing thatthe ruling concerns a limited set of lyrics and doesn’t affect millions of daily users in Germany.

Appeals could refine the reasoning or remedy the scope, but for now, the takeaway is clear: courts can treat model memorization as a form of reproduction.

That framing pressures AI firms to harden de-duplication, anti-memorization, and output filtering across European deployments.

Gavel on desk with judge working at courtroom.

Depositphotos

Setting a European precedent with global echoes

The judgment is being cast as a landmark European ruling. Even if not binding outside Germany, it provides persuasive authority for parallel suits against image, music, and text models.

Regulators and courts may cite Munich’s approach when interpreting exceptions, text-and-data-mining rules, and fair-use analogs.

Practically, it accelerates licensing talks across collecting societies, raising the cost and compliance requirements of training frontier models.

Depositphotos

What this means for AI training data

Developers now face a sharper legal distinction: using copyrighted works for training may be treated as reproduction when models memorize and can regenerate protected expression.

That nudges companies toward licensed datasets, aggressive deduping, and retrieval-based systems that avoid verbatim outputs.

It also elevates documentation duties, as proving the absence of copyrighted material or demonstrating licensed inclusion could become a table-stakes requirement in European markets.

European flags flap in the wind outside EU headquarters.

Ale_Mi/Depositphotos

How this interacts with EU policy trends

The ruling arrives as the EU finalizes enforcement of the I Act and continues debates on text-and-data-mining exceptions.

Courts are now adding case law that tightens expectations on transparency and licensing. Together, regulation and litigation form a pincer: disclose enough about training, avoid reproducing protected content, or obtain licenses.

That interaction may accelerate the development of standardized industry frameworks in Europe more quickly than in the US.

Want to see how other nations are tightening control on AI content? Read about how Japan is aiming to open up OpenAI over anime rights here.

Depositphotos

What to watch in the appeal and beyond?

On appeal, watch how judges parse memorization versus transformation, and whether injunctive relief forces technical disclosures.

Industry-wide, look for multi-territory deals with collecting societies, product changes to lyrics handling, and spillover cases involving images and code.

However it ends, Munich has turned a debate into a doctrine, pushing AI from “move fast” into “license and verify” in one of the world’s most consequential markets.

Interested in how creators are responding to AI’s rapid evolution? See why some are now inviting OpenAI to bring their characters to life in Sora here.

What do you think about OpenAI facing legal issues in German courts due to copyrighting AI-generated music? Please share your thoughts and drop a comment.