Can AI-Generated Music Be Copyrighted? Authorship and Risks

Music generated entirely by an AI tool cannot be copyrighted in the United States. When a person uses AI as one part of a larger creative process and contributes their own original expression — writing lyrics, performing vocals, arranging and mixing — the human-authored portions can be registered and protected, but the AI-generated material inside the same track cannot. So the honest answer to whether AI-generated music can be copyrighted is: only the human parts, and only if there are human parts.

Why Human Authorship Is Required

Federal copyright law protects original creative works fixed in a lasting form, and musical works and sound recordings are both eligible categories.¹ But protection has always required a person at the origin of the work. The Copyright Office has said since the 1970s that a work must “owe its origin to a human being” to be registered.²

In 2025, that principle became binding case law. In Thaler v. Perlmutter, the D.C. Circuit Court of Appeals held that an AI system “cannot be the recognized author of a copyrighted work because the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being.” The court noted that the structure of copyright — duration measured by an author’s lifespan, inheritance to a widow or children, signatures required to transfer ownership — only makes sense if the author is a person.³

Purely AI-Generated Tracks Have No Copyright

If you type a prompt into an AI music generator and it returns a finished song, that song has no copyright protection. The Copyright Office’s 2023 registration guidance is direct: when an AI receives a prompt and produces a complex musical work, the AI determines the expressive elements, not the user. The output “is not the product of human authorship” and “is not protected by copyright.”⁴

The consequence is real. Uncopyrightable works effectively sit in the public domain. Anyone can copy, remix, perform, or sell them. If you upload a purely AI-generated track to a streaming platform, you have no legal mechanism to stop someone else from distributing the same audio or building it into their own project.²

Detailed Prompts Don’t Give You Authorship

This is where most users misread the law. Writing a long, specific prompt describing genre, tempo, mood, instrumentation, and structure does not make you the author of what the AI produces. The Copyright Office’s 2025 Copyrightability Report concluded that “prompts alone do not provide sufficient human control to make users of an AI system the authors of the output.” Prompts convey ideas, and ideas are not copyrightable. What matters is control over the final expression, and a prompt doesn’t provide it.⁵

Iterative prompting doesn’t change the outcome. Tweaking a prompt over and over until you like the result is, in the Copyright Office’s phrase, “re-rolling the dice”: you’re selecting from what the AI generates, not authoring the expression. Time and effort spent refining prompts do not matter, because copyright protects original creative expression, not hard work.⁵

When AI-Assisted Music Qualifies for Copyright

Protection becomes available when a human actively shapes the final creative expression. The Copyright Office’s guidance recognizes two main paths: creatively selecting and arranging AI-generated elements, or modifying AI-generated material so substantially that the modifications themselves meet the standard for copyright.⁴

For music, that can look like:

Writing original lyrics and performing them over an AI-generated instrumental.
Recording live instruments alongside AI-produced tracks.
Substantially rearranging AI-generated sections into a new structure.
Mixing and producing the final recording with real creative decisions about sound design.

The test the Copyright Office applies is whether the human “had creative control over the work’s expression and actually formed the traditional elements of authorship.”⁴

You Own the Human Parts, Not the AI Parts

Even when your AI-assisted work qualifies, protection covers only what you created. The 2023 guidance is explicit: copyright in a mixed work “will only protect the human-authored aspects of the work,” and those protections are “independent of” and “do not affect” the copyright status of the AI-generated portions.⁶ The Copyright Office applied that logic to Kristina Kashtanova’s graphic novel Zarya of the Dawn: it registered her text and her creative selection and arrangement, but excluded the individual Midjourney-generated images from protection.⁷ Translate that to music: if you write lyrics and record vocals over an AI-generated beat, you own the lyrics and the performance. The beat itself stays unprotected.

AI Cannot Be a Co-Author

Naming the AI as a collaborator doesn’t rescue the AI-generated portion. The Copyright Office’s 2025 report confirmed that AI systems cannot be joint authors: they aren’t human, cannot form the intent required for a joint work, and cannot enter into binding contracts.⁵ Thaler reached the same conclusion through the statute itself.³

Registering an AI-Assisted Song

If your work mixes human and AI contributions, the Copyright Office requires the Standard Application (not the Single Application), and your filing must accurately describe which parts a human created and which parts the AI produced.⁶

In the “Author Created” field, describe what the human contributed. Examples: “lyrics,” “vocal performance,” “arrangement and mixing of AI-generated instrumental tracks.”
In the “Limitation of the Claim” section, under “Material Excluded,” describe the AI-generated portions. Example: “instrumental track generated by artificial intelligence.”
Use the “Note to CO” field to explain how you used AI and what creative decisions you made, if useful.

AI-generated content that is trivial — what the Copyright Office calls “de minimis” — doesn’t need to be disclaimed. Anything more substantial must be excluded from the claim.⁶ Leaving AI material out of your disclosure can jeopardize the whole registration if the Copyright Office later determines the application misrepresented what a human authored.

Other Legal Risks That Survive the Copyright Question

Copyright is not the only issue when you release AI-assisted music. A few risks apply even when your own authorship claim is solid.

Your AI Output May Resemble Someone Else’s Song

AI music models are trained on large catalogs of existing recordings, and outputs can end up sounding recognizably close to protected works. For compositions — melody, harmony, structure — the infringement test asks whether an average listener would find the similarity recognizable. For sound recordings, federal law draws a narrower line: copyright in a sound recording does not stop someone from independently creating new sounds that imitate the original. The statute says making “another sound recording that consists entirely of an independent fixation of other sounds” is not infringement, “even though such sounds imitate or simulate those in the copyrighted sound recording.”⁸ Whether AI-generated audio that closely mimics a real recording counts as “independent fixation” has not been settled by the courts. Separately, whether training AI models on copyrighted music is itself infringement is being litigated by major record labels against AI music companies, without a definitive answer yet.

Cloning a Real Artist’s Voice

Generating a song in a specific artist’s voice raises a separate risk beyond copyright. Roughly two-thirds of states recognize a right of publicity, which lets a person control commercial use of their identity, including their voice. Courts have found liability even when an imitation was not exact. There is no comprehensive federal right of publicity law, though the NO FAKES Act, which would create a federal right to control the use of your voice and visual likeness, was reintroduced in the Senate in April 2025.⁹ As of early 2026, the bill had not been enacted. Until it is, protection is a patchwork of state laws, and the safest path is to avoid generating output that could be mistaken for a real performer’s actual voice.

Platform Terms Are a Separate Contract

The AI service’s terms of use control what you can do with its output regardless of the copyright answer. Some platforms grant broad commercial licenses; others retain rights or restrict commercial distribution. Those are contract terms binding you as a user, not copyright rights, and they don’t help you stop third parties who obtain the same content elsewhere. Read the terms before you distribute anything commercially, make sure your human contribution is substantial enough to support a copyright claim, and register the work with accurate disclosures. The gap between “I generated this” and “I own this” is wider than most people expect.

1
Office of the Law Revision Counsel. 17 USC 102 – Subject Matter of Copyright In General
2
U.S. Copyright Office. Compendium Chapter 300 – Copyrightable Authorship What Can Be Registered
3
U.S. Court of Appeals for the D.C. Circuit. Thaler v Perlmutter
4
Federal Register. Copyright Registration Guidance – Works Containing Material Generated by Artificial Intelligence
5
U.S. Copyright Office. Copyright and Artificial Intelligence Part 2 Copyrightability Report
6
U.S. Copyright Office. Copyright Registration Guidance – Works Containing Material Generated by Artificial Intelligence
7
U.S. Copyright Office. Zarya of the Dawn Registration Decision
8
Office of the Law Revision Counsel. 17 USC 114 – Scope of Exclusive Rights in Sound Recordings
9
U.S. Congress. S 1367 – NO FAKES Act of 2025