Copyright Ownership Of AI-Generated Audio Compositions And Experimental Soundscapes.
📌 1) Thaler v. Perlmutter (U.S. Court of Appeals, D.C. Cir. – 2025)
Background:
Dr. Stephen Thaler created an AI called the Creativity Machine which generated visual art. He applied for copyright registration, claiming ownership.
Holding:
The U.S. Court of Appeals for the D.C. Circuit held that an AI itself cannot be an author under U.S. copyright law because the Copyright Act requires human authorship at the moment of creation. An output created entirely by a machine without creative contribution from a human cannot be copyrighted.
Why This Matters for AI-Generated Music:
Courts extend this logic across expressive media — music, soundscapes, compositions and any creative output. If a generative model produces an entire piece of music without meaningful human intervention, in the U.S., there is no human author to claim ownership. That output typically falls into the public domain or is simply unprotected by copyright.
This establishes a core principle:
Only human-authored or human-edited works can be copyrighted under current U.S. law.
📌 2) GEMA v. OpenAI (Regional Court of Munich, Germany – 2025)
Background:
GEMA — Germany’s major music rights society — sued OpenAI alleging its large language models were trained on copyrighted musical lyrics without permission and could reproduce them.
Ruling:
The Munich court found that the mere act of training an AI on copyrighted material without licences can constitute copyright infringement, including the internal memorization of protected content and its reproduction in outputs.
Key Legal Insight:
The court recognized that even if the copyrighted text is not reproduced in a human-readable manner inside an AI’s training data, its retention can violate reproduction rights under copyright law.
Applied to music, this means: training generative music models on unlicensed songs can itself be infringement — regardless of whether the generated audio exactly replicates any existing track.
Implications for AI Sound Generators:
This decision is significant because European copyright law often focuses on the act of reproducing protected material in any form. If an AI model’s training process qualifies as reproduction, developers and operators can be liable. This affects AI models that generate compositions by consuming huge music catalogues.
📌 3) UMG v. Udio and Parallel Suno Lawsuits (U.S. Federal Courts – 2024-2025)
Background:
Major U.S. record labels (Universal Music Group, Sony Music, Warner Music) plus RIAA filed lawsuits against AI music platforms Udio and Suno claiming they trained their generative models on copyrighted recordings without permission and produced infringing outputs. These cases are ongoing but very instructive.
Key Claims by Plaintiffs:
Using record companies’ sound recordings without licences to train AI music generators is copyright infringement.
Generated music files may closely resemble protected works filtered by prompts, causing direct and derivative infringement.
Developments:
UMG and Udio reached a settlement and licensing framework, where UMG artists and songwriters get compensated and transfer rights for licensed AI training/outputs.
Suno also agreed to licensing arrangements with major music groups to comply with copyright standards.
Importance:
These cases signal a shift from pure litigation to licensing regimes for AI music — suggesting that industries see generative music tools as legitimate but requiring contractual copyright licensing of training data and output rights.
📌 4) Zarya of the Dawn – U.S. Copyright Office Ruling (2023/2025 Precedents)
Background:
A graphic novel consisting of images generated by Midjourney (an AI tool) and arranged manually. The U.S. Copyright Office initially granted registration but later revoked copyright protection for the AI-generated illustrations themselves. Human-written text and layout remained protected.
Principle Applied to Music:
Although not a case in a traditional court, the Copyright Office’s action clarifies enforcement policy:
AI-generated elements do not qualify for copyright unless there is substantial human creative input.
Simple prompts or generation without artistic decision-making is insufficient.
This policy guides registries worldwide about what constitutes protectable authorship in generative work — music included.
📌 5) Li v. Liu (Beijing Internet Court – China, 2023)
Facts:
A Chinese artist used an AI tool to create an image, choosing prompts, settings, and refining generations.
Holding:
The Beijing Internet Court granted copyright, finding that human choice in prompting and editing reflects enough personal intellectual investment. The copyright was owned by the human user, not the AI tool.
Relevance to Soundscapes & Music:
This decision is often referenced in comparative law discussions:
AI does not obtain rights.
Human curated prompts, selection, tweaks and creative direction can establish authorship.
For AI-assisted music compositions, this means that a human who curates outputs, modifies structure or deliberately shapes the work might be the real author, even if generative models did much of the raw creation.
🔎 Synthesis: What Courts Are Really Saying
🧠 Authorship
AI is not a legal author under current copyright frameworks (especially in the U.S. due to Thaler v. Perlmutter).
Only human creators can own copyright, unless national law changes expressly include AI.
🛠 Infringement Liability
Training on copyrighted music without licences may itself be infringement (GEMA v. OpenAI; UMG v. Suno/Udio).
Outputs that are too close to existing works can trigger infringement claims.
🎼 Human Creative Input Matters
Courts and copyright offices require evidence that the human did more than hit “generate”: selecting, editing, restructuring, harmonizing, or otherwise shaping the work is what makes it protectable.
📜 Licensing as a Commercial Solution
Lawsuits against generative music startups often end with licensing agreements — this reflects industry practice: AI tools must respect the existing copyright ecosystem to be sustainable.
📌 Practical Takeaways for AI-Generated Music Creators
If the human contribution is minimal or nonexistent, the AI output likely cannot be copyrighted under current law.
If a human shapes, edits, arranges, or selects AI material significantly, that human may claim authorship of the resulting work.
Use licensed datasets for training generative models or obtain licences where training uses real copyrighted material.
Always evaluate whether outputs resemble protected works too closely — similarity tests may trigger infringement liability.
RELATED Blog
comments