Who Owns a Hybrid Record? The AI Copyright Walkthrough Nobody Gives You
You can copyright a record made with AI — thousands of creators already have. But the application punishes guesswork. Here is the field-by-field walkthrough, translated into working producer language.
The question inevitably lands in every producer’s inbox, usually at 1:00 a.m., right after the master is bounced: “I used AI on this — can I even copyright it?”
The answer is yes, with strict conditions. And those conditions are exactly where catalogs get quietly wrecked. Thousands of hybrid human-AI works have successfully registered under the U.S. Copyright Office’s framework. The ones that fail always share one of two fatal errors: attempting to claim machine output as human authorship, or attempting to hide the machine’s involvement entirely. Both errors are completely avoidable with twenty minutes of focused understanding. This is those twenty minutes.
The legal framework in three sentences
Copyright protects human authorship exclusively; machine-generated expression cannot be registered, and AI prompting — no matter how elaborate — is considered instruction, not authorship. To register a hybrid work, you claim the human contributions while formally disclosing and excluding the machine’s output, whenever the AI material is more than de minimis. Omitting this disclosure is not a clever shortcut; a registration obtained with a knowingly inaccurate application can be invalidated later — meaning the certificate you’d need in a catalog dispute is exactly what dishonesty destroys.
The two copyrights inside your record
Before you touch the application, split your record exactly the way the law does. Every track carries two copyrights: the composition (PA) — the song as written: melody, lyrics, structure — and the sound recording (SR) — the specifically captured audio. Your AI use likely touches them differently.
Here’s the plainest way to hold this distinction, because everything in a hybrid registration depends on it: your record is two properties, not one — the song, and the recording of the song. Think of it as a recipe and a dish. The composition (the PA side — “publishing”) is the recipe: melody, lyrics, chords, structure — the thing that exists even on a napkin, before anyone presses record. The sound recording (the SR side — the “master”) is the dish as actually cooked: these specific sounds, this performance, this audio. You already know they’re separate even if you’ve never said it this way — it’s why cover versions exist at all: when another artist covers your song, they use your recipe but cook their own dish. Two properties, two possible owners, two different sets of questions on every form.
Now run the machine through it and watch the two properties split:
Scenario one: you wrote every note and lyric, then generated the backing audio. Your recipe is one hundred percent yours — the composition claim is strong, full stop. But the dish? A machine cooked most of it. The actual sounds in that master are generated audio, and your sound-recording claim covers only what’s genuinely human in the audio — your vocal if you sang it, your selection and arrangement of the material — with the generated audio disclosed and excluded. Strong song, thin recording. That’s not a failure; it’s the honest shape of what you made.
Scenario two: the machine sketched the arrangement, and you re-performed every sound — sang every line, played every part, replaced every generated element until nothing machine-made survived in the audio. Now the dish is entirely your cooking: the sound-recording claim is fully human and strong, and where you authored or genuinely commanded the underlying material, the composition claim stands alongside it. Both properties, both yours.
And here is the single most common way hybrid registrations go wrong: producers answer recording questions with their songwriting reality. Picture it concretely. A producer who wrote every word and melody himself — genuinely, at the piano, with a pencil — generates the instrumental with AI, and on the form answers the AI-material questions “none,” because in his mind “I wrote this whole song myself” is simply true. It is true — about the recipe. But the form was asking about the dish, and the dish is mostly machine-cooked audio. He hasn’t lied about his songwriting; he’s misdescribed his master — and the registration now misrepresents the very recording it claims to cover. The reverse error happens too: a producer who performed every single sound panics at the AI questions because a machine helped brainstorm the sketch he later rewrote — under-claiming a recording that is entirely his. The cure for both is one discipline, and you already keep it if you keep the Machine Log: map exactly which layer the machine touched — song or sounds, recipe or dish — before you open the form. Answer each property’s questions about that property, and the hybrid registration stops being a minefield and becomes what it was always supposed to be: a description.
The application, field by field
The Standard Application’s decisive fields for a hybrid work:
“Author Created” — state your human contributions in established terms: “Original lyrics, music composition, and human vocal performance” — whatever is strictly true of your record. Never list an AI system as an author or co-author. The machine is not a collaborator in the eyes of the law; it is a tool whose output you are about to explicitly exclude.
“Limitation of Claim → Material Excluded” — this is your disclosure. Check the applicable boxes and describe the machine material in the “Other” field: “[Description of the relevant audio] generated by artificial intelligence.” Plain, specific, unpanicked.
“Limitation of Claim → New Material Included” — describe the human authorship you are claiming, matching your “Author Created” language: “Selection, coordination, and arrangement of human and non-human audio, plus original lyrics.”
Consider a track with original human lyrics and vocals over a generated instrumental: the PA claim covers lyrics and composition while excluding the AI instrumental elements; the SR claim covers the human vocal performance and mix while excluding the AI-generated sound-recording material. Copyright Office examiners assess case by case whether the human contribution is perceptible in the deposited work — the real standard hiding under all the paperwork: they have to be able to find you in the record.
The part the form cannot see
Everything above assumes a fact pattern you can back up. That is the secondary job of your Provenance Dossier →: the session histories, raw takes, and generation logs that document exactly where your human authorship lives aren’t just platform armor — they are what makes your Limitation of Claim a description of reality rather than a hopeful guess.
Register honestly, document contemporaneously, and the certificate becomes exactly what it is supposed to be: the deed to something you can prove you built.
A necessary caveat: this walkthrough is education from a working producer’s practice, not legal advice. AI copyright doctrine is evolving rapidly; for records with real money attached, put an entertainment attorney’s eyes on the actual application.