Rights & policy
AI Music vs Sampling: Where Does Influence Become Copying?
Mobb Deep and Daft Punk turned existing recordings into new music. AI training raises a different set of rights questions. Here is what the comparison proves—and what it cannot settle.
AI-assisted
AI helped research and draft this analysis, developed alongside Nik McFly’s manifesto. The linked sources distinguish production history, copyright guidance and a specific court decision.
A few seconds of an old recording can become the sound people remember from a new song. That is the appeal of a good sample breakdown: the source plays, the producer changes it, and a familiar hook suddenly appears.
AI music brings a harder question into that familiar scene. If working with earlier music can be creative, what makes a particular use of it copying? The answer changes depending on whether the discussion concerns a recording, a composition, model training or the conduct of the person releasing a track.
Editorial disclosure: This analysis accompanies a manifesto by Nik McFly, the founder and editor of AI Music Events, who uses AI in his music business. It is a companion explanation from the same publisher, not independent coverage of his position.
What Havoc actually did with “Jessica”
In his Song Exploder interview, Havoc explains how Herbie Hancock’s “Jessica” became part of Mobb Deep’s “Shook Ones, Pt. II.” He experimented with a roughly two-second passage in a keyboard sampler, changed its pitch and filtered a piano note into a bass line. A sound from a Quincy Jones record helped mark the chorus.
Those decisions changed the role of the recordings. A piece of piano became material for a different mood, rhythm and song. The listener could recognise the new work without identifying the old one.
The example demonstrates creative transformation. It does not, on its own, establish the permission or payment terms for those samples.
Daft Punk makes the distinction clearer
Tracklib’s breakdown of “One More Time” traces its source to Eddie Johns’ “More Spell On You.” The demonstration makes the rearrangement audible. Tracklib also reports that the sample had been cleared, despite years of speculation about its use.
That detail matters. A producer can make a creative contribution and still need permission for the material being transformed. Clearance and creativity answer different questions.
The comparison supports a limited but important argument: using existing material does not automatically erase the creativity of the new work. It cannot turn every use of existing material into an authorised one.
Four things that get called “copying”
| Practice | What to examine |
|---|---|
| Musical influence | Shared styles and techniques, and whether any specific protected expression is reproduced |
| Sampling | The actual recording segment reused, the underlying composition and the terms that cover the use |
| Model training | How the developer obtained and processed its data, and which permissions or legal exceptions apply |
| A generated sound-alike | Which elements of a particular work appear in the output and who is responsible for that use |
The U.S. Copyright Office’s guidance for musicians separates rights in the composition from rights in the recording. Using another artist’s work generally requires permission, a suitable licence, public-domain status or an applicable legal exception. There is no universal number of seconds that makes an otherwise unauthorised sample safe.
A training dispute can concern copying during data collection or training even if a particular released song does not resemble a source. Conversely, calling something a generated result does not establish that it is free of copied expression.
What the German Suno decision actually addressed
On 31 July 2026, the Munich Regional Court I largely upheld GEMA’s claims against Suno concerning six musical works. Its public account of the decision addressed copying during training, memorisation in models and recognisable elements in generated outputs. The court rejected Suno’s fair-use defence on those facts.
For the outputs at issue, the court attributed responsibility to Suno rather than the users who entered the prompts. The release stated that the judgment was not final when announced.
This was a decision about specified works, models and conduct. It was neither a finding that every Suno user had stolen a song nor a promise that users can never be responsible for what they publish.
Where the argument becomes a choice about culture
McFly’s manifesto makes a broader claim: access to shared culture helps people learn and create, and machine learning can become part of that exchange. He describes contributing to learning as a price of the opportunities the internet provides.
That is a position about how the exchange should work. Internet publication does not itself provide an AI company with a licence. Whether creators should receive payment, how permission should be obtained and who benefits from a model remain substantive questions.
The strongest version of the sampling comparison starts with a fact about creativity: transforming earlier work can produce something new. The strongest objection asks about the conditions of that transformation—permission, control, compensation and the result that reaches the listener.
Those questions can be asked of both a producer and an AI company. They do not require pretending that a sampler and a training system perform the same operation.
A useful accusation has to name the act
Someone challenging a release can identify the source work and the elements they believe were copied. Someone challenging training can identify the works used, the developer’s conduct and the rights they believe were infringed. A criticism of market power can explain who gains, who loses and what rule would improve the outcome.
Each is more informative than treating influence, training and a copied song as one thing.
Read McFly’s full manifesto, or explore the AI music rights and copyright guide for a separate practical overview. The examples above explain the debate; they do not determine the rights in a reader’s own release.