AI Music Training Found Infringing in Europe — What the GEMA v. Suno Ruling Means for Indie Artists

A Munich court ruled on July 31, 2026 that Suno, one of the most popular AI music generators, infringed copyright by training its model on protected songs without authorization. The decision marks the first time a European court has held an AI company liable for using copyrighted music as training data — and its reasoning has immediate consequences for indie artists worldwide.

What the Court Found

The Munich Regional Court sided with GEMA, Germany’s music collecting society, finding that Suno’s AI model reproduced six well-known compositions — including “Rasputin,” “Atemlos durch die Nacht,” “Big in Japan,” “Forever Young,” and “Mambo No. 5” — stored within its model weights.

The court’s central holding rests on what legal scholars call the memorization doctrine: when an AI model stores copyrighted works in its parameters in a form that allows it to reproduce those works in its outputs, that storage constitutes unauthorized reproduction under copyright law. This is distinct from analytical processing — the court drew a sharp line between extracting information from a work (potentially permitted) and memorizing the work itself (infringing).

Crucially, the court held Suno liable, not the user. Because Suno selected the training data, designed the model architecture, and controlled how outputs were generated, the company bore responsibility for any infringing reproductions its model produced.

Why the EU Text-and-Data-Mining Exception Didn’t Help

Suno argued that the EU’s text-and-data-mining (TDM) exception under §44b of the German Copyright Act — which permits automated analytical extraction of copyrighted works for research purposes — shielded its training activities. The court rejected this argument.

The TDM exception covers extraction for analytical purposes: identifying patterns, counting occurrences, extracting metadata. It does not cover storing complete or near-complete copies of copyrighted works in model weights in a form that enables reproduction. The court found that Suno’s training went beyond analysis into memorization, placing it outside the scope of the exception.

This distinction matters because the TDM exception was explicitly designed to facilitate data science and research, not to provide a blanket license for building commercial products that compete with the works they were trained on.

Fair Use Rejected Under U.S. Law

In a notable move, the German court also evaluated Suno’s defense under U.S. fair use doctrine, since some training occurred in the United States. The court rejected fair use on all four factors:

1. Purpose and character: Suno’s use was commercial and non-transformative — it built a product that generates music in the same market as the training data.
2. Nature of the original works: The compositions were creative, published works deserving strong protection.
3. Amount used: Suno copied entire works, not excerpts.
4. Market effect: Suno’s outputs could serve as substitutes for the original recordings.

The court distinguished recent U.S. rulings like Bartz v. Anthropic and Kadrey v. Meta, where courts found AI training transformative, by noting that those cases did not involve outputs substantially similar to the training data. Suno’s model, by contrast, could produce outputs that closely resembled the songs it was trained on.

Stream-Ripping: Bad-Faith Data Acquisition

The court also found that Suno acquired its training data through “stream-ripping” — using tools to bypass YouTube’s Rolling Cipher protection system and extract audio streams. This was treated as evidence of bad-faith acquisition, undermining Suno’s argument that it had acted reasonably.

This finding is significant because it suggests that how training data is obtained matters, not just whether training is permitted. AI companies that circumvent technical protection measures to acquire training data face additional legal exposure beyond the training act itself.

What This Means for Indie Artists

Your Music Has Likely Been Scraped

The ruling confirms what many independent artists suspected: AI companies have been training on copyrighted music at scale, including music by unsigned and independent artists. Suno’s model was not limited to major-label catalog — it ingested broadly from available audio sources across the internet.

Individual Enforcement Is Difficult

While the ruling strengthens rightsholders’ legal position, enforcing those rights as an individual indie artist remains practically challenging. GEMA brought this case as a collecting society with the resources to litigate against a well-funded AI company. Most independent artists lack the legal budget for individual enforcement.

Collective action is the most realistic path. Organizations like GEMA (Germany), SACEM (France), PRS (UK), and ASCAP or BMI (U.S.) are better positioned to negotiate licensing frameworks and pursue litigation on behalf of their members. Indie artists should consider:

• Joining a collecting society if they haven’t already
• Monitoring opt-out mechanisms under the EU Copyright Directive
• Supporting industry efforts to establish licensing frameworks for AI training

Licensing Is Emerging — But Not Yet Available to Most Indies

Suno has signed licensing agreements with Warner Music Group, BMG, and Believe since late 2025. These deals acknowledge that a licensing market for AI training data exists — a direct contradiction of the position that training is fair use and requires no payment.

However, these deals cover major and large independent catalogs. There is currently no standard mechanism for indie artists to license their music to AI companies or to receive compensation when their music has already been used. The GEMA ruling may accelerate the creation of such mechanisms, but they do not yet exist in a practical form.

The EU AI Act Adds a Compliance Layer

As of August 2, 2026 — just two days after the GEMA ruling — the EU AI Act’s transparency provisions became enforceable. These require:

• Machine-readable watermarks on AI-generated audio (existing models have until December 2, 2026 to comply)
• Training data summaries from general-purpose AI model providers
• Copyright compliance policies that respect rightsholder opt-outs

Together with the GEMA ruling, the EU AI Act creates a regulatory environment where AI music companies operating in Europe face both legal liability for past training and compliance obligations for future operations.

The Extraterritorial Reach

One of the ruling’s most far-reaching aspects is its assertion of extraterritorial jurisdiction. The Munich court found Suno liable for training that occurred in the United States, reasoning that the effects of that training were felt in Germany through Suno’s service to German users.

This means AI companies cannot avoid European copyright law simply by training outside the EU. If they serve EU users, they may be subject to EU copyright standards. For indie artists, this extends the practical protection of the GEMA ruling beyond German borders to potentially any market where the AI company operates.

What Comes Next

Suno may appeal the ruling, and the case could take years to reach final resolution. The court made its order provisionally enforceable against security, so Suno must stop reproducing the six compositions at issue and disclose its revenues for damages calculation while any appeal is pending.

The legal pressure intensified further on September 18, 2026, when Universal Music Group and Sony Music filed a second lawsuit against Suno in U.S. federal court, this time targeting its v6 model suite launched on September 9, 2026. The complaint alleges that v6 was built using “knowledge distillation” from earlier models trained on unlicensed recordings, arguing that the new models embody the “accumulated learnings” of infringing predecessors. The labels describe v6 as “the fruit of the same poisoned tree” and seek damages for 60,202 specific recordings at up to $150,000 per work for willful infringement. This second lawsuit tests whether AI companies can launder allegedly infringing training data by rebuilding models on outputs of earlier infringing models — a question with implications far beyond music.

More broadly, the ruling has set a precedent that other courts may follow. The CJEU (Court of Justice of the European Union) has a pending preliminary reference (Case C-250/25) on whether LLM training constitutes copyright-relevant reproduction — its decision could harmonize this approach across all EU member states.

In the United States, the major-label lawsuits against Suno and Udio continue, with a fair use ruling not expected until 2027. Independent artists are pressing a separate case: on August 20, 2026 the federal court in Boston denied most of Suno’s motion to dismiss a putative class action, allowing copyright and DMCA circumvention claims — including the theory that Suno bypassed YouTube’s rolling cipher to obtain training data — to move forward. That order decides nothing about whether Suno actually infringed. The GEMA decision does not bind U.S. courts, but it may influence the legal analysis, particularly on the question of market harm. Note that the ruling concerns training, not who owns what the models output — that question is answered in Is AI music legal? Who owns AI-generated music.

Practical Steps for Indie Artists

• Register with a collecting society if you haven’t already — collective enforcement is the most viable path
• Document your creative process — detailed records strengthen any future claim
• Monitor AI platform terms — as companies like Suno update their terms in response to litigation, check what rights you retain
• Use opt-out mechanisms where available — the EU Copyright Directive provides for TDM opt-outs, and tools are emerging to exercise them
• Check distributor policies — some distributors are beginning to address AI training in their terms
• Support industry coalitions — organizations like the Artist Rights Alliance and IFPI are advocating for creator protections

Common Misconception

“The GEMA ruling only protects German artists.” The ruling protects all rightsholders whose works were used for training, regardless of nationality. GEMA brought the case on behalf of its members and affiliated international rightsholders. The court’s reasoning about memorization, TDM exceptions, and fair use applies broadly — it does not depend on the artist’s country of origin. That said, enforcing these rights as an individual outside a collecting society structure remains difficult.

When to Ask a Lawyer

• If you have evidence that a specific AI model reproduced your music in its outputs
• If you are considering joining or forming a collective licensing initiative
• If you are negotiating with a distributor or platform that is changing its AI-related terms
• If you want to understand your rights under the EU Copyright Directive’s opt-out provisions
• If you have been approached by an AI company seeking to license your catalog

Sources

1. Munich Regional Court, GEMA v. Suno, judgment of July 31, 2026 — First-instance ruling finding AI music training constitutes copyright infringement under German law
2. EU AI Act, Regulation (EU) 2024/1689, Articles 50 and 53 — Transparency and copyright compliance obligations for general-purpose AI models
3. Directive (EU) 2019/790, Article 4 — Text-and-data-mining exception and rightsholder opt-out mechanism
4. U.S. Copyright Office, “Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence,” Federal Register, Feb. 2023
5. CJEU Case C-250/25 (Like Company) — Pending preliminary reference on LLM training and copyright reproduction
6. Justice v. Suno, Inc., No. 1:25-cv-11739 (D. Mass., memorandum and order on motion to dismiss, Aug. 20, 2026) — copyright and DMCA claims allowed to proceed

Frequently Asked Questions

Can AI companies train on copyrighted music without permission?

A German court ruled in July 2026 that training AI models on copyrighted music without permission constitutes copyright infringement. The ruling applies to any AI company serving EU users, regardless of where training occurs.

What was the GEMA v. Suno court ruling about?

The Munich Regional Court found that Suno's AI music generator reproduced copyrighted songs stored in its model weights, rejecting both the EU text-and-data-mining exception and U.S. fair use as defenses.

Does the GEMA ruling apply to indie artists or only major labels?

The ruling protects all rightsholders whose works were used for training, including independent artists. However, collective enforcement through organizations like GEMA remains the most practical path for individual creators.