News · legal — By WATMM Desk —
Can you copyright a song made with AI? The US Copyright Office rules in plain English
In the US, a prompt on its own does not make you the author of what an AI generates. Lyrics, melodies, performances, edits and creative arrangement you made yourself can still be registered, as long as you claim your part and disclose the AI part on the form.
Key facts
- The rule: the US Copyright Office's January 2025 report concludes that, with current generally available technology, prompts alone do not give enough control over the output to make the user its author.
- What is protected: your own expression that is perceptible in the output, your creative selection, coordination or arrangement of material, and creative changes you make to AI output, judged case by case.
- On the form: use the Standard Application, describe the human authorship under "Author Created", and exclude AI-generated material that is more than de minimis under Limitation of the Claim, Material Excluded, Other.
If you make music with AI tools and want to own the result in the US, the Copyright Office has set out where the line sits. The Part 2 report on copyrightability, published on 29 January 2025, gives the reasoning. The 2023 registration guidance tells you how to fill in the form. The Office's AI page lists no newer registration guidance as of our check on 23 October 2026. This is an explainer, not legal advice.
What does the Copyright Office actually say?
Copyright protects only human authorship. The Office's view is that, with current generally available technology, prompts alone don't give you enough control over the output to make you its author. A prompt works like an instruction to a commissioned artist: you say what you want, and the system decides how it sounds. Rewriting the prompt and generating again doesn't change that. The report calls it "re-rolling" the dice, and says the time or effort you put in is irrelevant.
Three kinds of human contribution can still be protected, each judged case by case:
- Your expressive input. If you feed in something you made, such as a melody, a vocal or a demo, and it is still perceptible in the output, you are the author of at least that part. The protection covers your perceptible expression, much as it would for a derivative work.
- Selection, coordination and arrangement. Arranging material in a creative way can be protected as a whole, though not the AI-generated pieces on their own. The report notes that selecting just two or three elements is generally not enough.
- Modification. Changes to AI output that meet the usual originality standard are protected.
Using AI as a tool doesn't take protection away. The report's music example is a Randy Travis recording. An AI vocal model turned another singer's human performance into a track in Travis's voice, and the Office registered the sound recording because AI was used as a tool rather than to generate the expression.
The courts agree on the basic rule. In March 2025 the D.C. Circuit upheld the refusal of a registration that named a machine as the sole author, holding that the Copyright Act requires a human author. It said arguments over how much AI is too much were not part of that case. The Supreme Court denied review on 2 March 2026. So the human-authorship rule stands, and where exactly the line falls is still up to the Office, case by case.
Four producer scenarios
These examples are our reading of the guidance, not rulings. The Office decides each claim on its facts.
- Typed a prompt, released the output. There is no human authorship to claim in the music or the recording. Nothing in the guidance stops you releasing it, but without copyright you would have little to stand on if someone copied it. It also affects your income. The Office has told the Mechanical Licensing Collective that owners of musical works without copyright aren't entitled to its blanket-licence royalties. It added that the collective may hold royalties while it investigates works that look AI-made, for example when someone claims an extraordinary number of songs in a short time.
- You wrote the lyrics and topline, AI made the backing. Claim the lyrics and melody. Exclude the AI-generated instrumental. If you sang the vocal yourself, that recorded performance is human authorship as well.
- You hummed a melody into a model and it rendered a full arrangement. Your melody is protected where it is clearly audible in the output. Whatever the model added is not. The report contrasts a registered image in which the artist's drawing stayed clearly visible with a refused one in which the human input didn't show through.
- You generated stems, then cut, re-recorded and rebuilt the track. This is your strongest claim: creative arrangement plus real modification. Even so, the AI material underneath stays unprotected.
How to fill in the form
The 2023 guidance gives the mechanics:
- Use the Standard Application. The guidance says works containing AI-generated material must use it. As of October 2026 the online fee is $65, compared with $45 for the Single Application; both rise on 11 November (see our fee coverage).
- Name yourself as author. Don't list the AI tool or its company as an author or co-author.
- "Author Created": briefly describe what the human did. The Office's model wording for arrangement is "Selection, coordination, and arrangement of [human content] created by the author and [AI content] generated by artificial intelligence."
- Limitation of the Claim, Material Excluded, Other: exclude any AI-generated material that is more than de minimis. Describe it as "[description] generated by artificial intelligence."
- Note to CO is optional. If you're unsure, a general statement that the work contains AI material is enough, and the Office will contact you.
- Composition and recording are separate works. Circular 56A lets you register both on one application if they are on the same recording and you own both. Disclose the AI part of each.
- Already filed without disclosing? If the application is still pending, contact the Public Information Office. If it has been registered, file a supplementary registration ($100 online).
Why bother registering
For a US work, you generally can't sue for infringement until the work is registered (17 U.S.C. § 411(a)). Statutory damages and attorney's fees are only available if you registered before the infringement began, or within three months of first publication (§ 412). Honesty on the form matters too. A court can disregard a registration if inaccurate information was included knowingly and the Office would have refused it had it known, and the guidance warns that the Office may cancel registrations where essential information was left out.
What we think
The rule is fairer to working musicians than it first sounds. It protects what you play, write and decide, and refuses to hand ownership of generic output to whoever typed a prompt. That holds back the flood of machine-made catalogue competing for the same royalty pools, and the MLC letter shows the Office applying the same logic to money, not just paperwork.
The cost falls on documentation. "Case by case" means the producer who can show what they did is the one who wins. A producer who only kept the bounce has nothing to show. Labels and publishers with legal teams will handle this easily, while bedroom producers are more likely to over-claim by mistake, and over-claiming is what puts a registration at risk. Save your stems, takes and project versions as you go. A workflow tool that keeps session history, Veena among them, makes that record easier to keep.
Watch the AI platforms' terms too. If a service says you "own" its outputs, we read that as the platform giving up its own claims, not as a grant of copyright. A contract can't create authorship that the law doesn't recognise.
Checklist before you file: list every AI tool and what it produced. Keep your own lyrics, melodies, recordings and edits. Choose the Standard Application. Write "Author Created" in plain terms. Exclude the AI material. Register within three months of release.
Sources
- Copyright and Artificial Intelligence, Part 2: Copyrightability, U.S. Copyright Office (primary source). Accessed 23 Oct 2026.
- Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence (88 Fed. Reg. 16190), U.S. Copyright Office (primary source). Accessed 23 Oct 2026.
- Copyright and Artificial Intelligence, U.S. Copyright Office (primary source). Accessed 23 Oct 2026.
- Royalty Eligibility of Musical Works Generated Using Artificial Intelligence Under the Music Modernization Act, U.S. Copyright Office (primary source). Accessed 23 Oct 2026.
- Thaler v. Perlmutter, No. 23-5233 (D.C. Cir. 2025), U.S. Court of Appeals for the D.C. Circuit (primary source). Accessed 23 Oct 2026.
- Docket No. 25-449, Thaler v. Perlmutter, Supreme Court of the United States (primary source). Accessed 23 Oct 2026.
- Copyright Registration of Musical Compositions and Sound Recordings (Circular 56A), U.S. Copyright Office (primary source). Accessed 23 Oct 2026.
- Fees, U.S. Copyright Office (primary source). Accessed 23 Oct 2026.
- 17 U.S. Code § 411, Legal Information Institute, Cornell Law School (primary source). Accessed 23 Oct 2026.
- 17 U.S. Code § 412, Legal Information Institute, Cornell Law School (primary source). Accessed 23 Oct 2026.
Researched with AI assistance, written and checked by WATMM Desk.
Published by the company behind Veena. Disclosures