1. What “Digital Royalties for AI Content” means in Germany

In Germany, “digital royalties” for AI content refers to payment obligations for using copyrighted works in AI systems, covering two stages:

(A) Training Stage (Input side)

• Using copyrighted works (music, text, images, code) to train AI models
• Issue: Is this “reproduction” under the German Copyright Act (UrhG)?

(B) Output Stage (Generation side)

• AI-generated outputs that reproduce or resemble protected works
• Issue: Is this an infringement or derivative use requiring payment?

Core Legal Principle in Germany

German copyright law is author-centric:

• Only humans can be “authors”
• Copyright protects expression, not ideas
• Any reproduction, adaptation, or public communication requires permission unless an exception applies

Relevant provisions:

• § 16 UrhG – reproduction right
• § 19a UrhG – making available to the public
• § 23 UrhG – adaptation/derivative works
• § 44b UrhG – Text and Data Mining (TDM exception)

2. Key Legal Shift: From “Free Data Use” → “Licensing Requirement”

Germany is currently shifting toward:

• AI training not automatically covered by TDM exception
• Strong push for licensing + royalty systems
• Recognition that AI models may “memorize” copyrighted works

This creates the foundation for:

AI-generated content royalties (AI licensing markets)

3. Major Case Laws (Germany + EU influence)

Below are 6+ key cases shaping AI royalties and copyright in Germany

CASE LAW 1: Munich Regional Court I (GEMA v OpenAI) – 2025

Decision

The court held that:

• AI models can “memorize” copyrighted works
• Both training storage + output reproduction = infringement

Key Holding:

• AI training is a reproduction under §16 UrhG
• Output that reproduces lyrics = infringement
• TDM exception does NOT apply

Impact on royalties:

✔ Strong basis for mandatory licensing for AI training
✔ Supports royalty-based AI content market

CASE LAW 2: Hamburg Regional Court (AI Training Allowed under TDM Exception) – 2024

Decision:

• AI training may be allowed under § 60d UrhG (research/TDM exception)

Key reasoning:

• Some AI training qualifies as:
  • “Automated analysis of data”
  • Non-expressive intermediate copying

Impact:

✔ Supports limited royalty-free AI training
❌ But conflict with Munich ruling creates legal uncertainty

CASE LAW 3: Federal Court of Justice (BGH) – “Inkasso-Programm” Principle (1985)

Principle established:

• Algorithms themselves are not copyrightable
• Protection applies only to expressive implementation

Why it matters for AI royalties:

• AI models are not “authors”
• Only outputs or training inputs matter
• Supports licensing of data, not algorithms

CASE LAW 4: EU Court of Justice (Infopaq Doctrine – EU influence applied in Germany)

Principle:

• Even small fragments of copyrighted works can be protected if:
  • They reflect author’s intellectual creation

Impact on AI:

• Training data fragments may still require licensing
• Strengthens argument for micro-royalty systems for datasets

CASE LAW 5: LAION / Image Dataset Case – Hamburg Regional Court (2024)

Decision:

• Scraping images for AI training can be lawful under certain conditions
• Especially when:
  • Data is publicly available
  • Used for research purposes

Impact:

✔ Supports open datasets
❌ But limited when commercial AI use is involved

CASE LAW 6: UK High Court (Stability AI v Getty Images – persuasive authority)

Decision:

• AI model training did NOT automatically constitute infringement in UK jurisdiction
• Focus on:
  • Where training occurred
  • Whether copying was stored or reproduced

Relevance to Germany:

• Often cited as counterweight to stricter German approach
• Highlights EU divergence in royalty enforcement models

CASE LAW 7: CJEU “TDM Directive Framework” (EU-wide binding law influencing Germany)

Principle:

• Text and Data Mining is allowed **only if:
  • lawful access exists
  • rights holders have not opted out**

Impact on royalties:

• Introduces opt-out licensing systems
• Basis for EU-wide collective AI licensing models

4. How Digital Royalties for AI are Expected to Work in Germany

Based on these cases, Germany is moving toward 3 royalty models:

(1) Dataset Licensing Model

AI companies pay:

• Publishers
• Music societies (like GEMA)
• Image agencies

💰 Royalties per dataset or per training volume

(2) Usage-Based Royalties

Creators get paid when:

• Their style/content is reproduced
• Their work influences AI outputs

💰 “Per-output” or “per-use” micro-royalties

(3) Collective Licensing System

Similar to radio/music royalties:

• AI companies pay a central collecting society
• Distributed to creators

Germany is particularly aligned with this model via organizations like:

• GEMA (music rights collective)

5. Key Legal Conflict in Germany

The central legal tension is:

✔ Pro-royalty position:

• AI training = reproduction → requires license
• Outputs can infringe even if not identical

❌ AI industry position:

• Training = statistical learning (not copying)
• Output = user-generated, not developer liability

6. Final Legal Status (2026 Outlook)

In Germany today:

• AI training is legally risky without licenses
• Courts are increasingly recognizing:
  • memorization = reproduction
  • outputs = potential infringement

Strong movement toward:

“AI royalty ecosystems” similar to music streaming rights

7. Conclusion

Germany is currently one of the strictest AI copyright jurisdictions in Europe, and its courts are actively shaping a system where:

• AI training is likely to become licensed and paid
• Outputs may trigger secondary royalties
• Collective rights organizations (like GEMA) will play a central role