Can I Copyright a Song I Made with AI?

User avatar placeholder
Written by Brian McBrearty

May 8, 2026

Yes, the portions you created yourself are protectable by copyright.

If your new track was half you, half AI, you still created the “I DID THAT” half. Copyright rules are changing, but your original expression will still be yours. This article will cover what the U.S. Copyright Office’s January 2025 Part 2 report on Copyrightability actually says, how to register AI-assisted songs, taking advantage of GRAM (Group Registration for Works on an Album of Music) for albums, especially in light of the proposed 2026 fee increases, and a little bit on the current state of the Suno and Udio litigations to the extent we should care.

The short answer again?

The parts you wrote are yours. The parts the AI wrote aren’t. The AI stuff kinda isn’t anybody’s. The Copyright Office’s Part 2 report put it this way, and it makes perfect sense to me: copyright protects the original expression in a work created by a human author, even if the work also includes AI-generated material; copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements. “The (copyright) Office has registered hundreds of works that incorporate AI-generated material, with the registration covering the human author’s contribution to the work.” And the Supreme Court reinforced it all on March 2, 2026, when it denied certiorari in Thaler v. Perlmutter, and preserved the D.C. Circuit’s holding that human authorship is a bedrock requirement. I don’t expect it to change.

You have to “Musicologize” your AI-assisted works.

It’s a ton like what Musicologize does with all music copyright-related questions, “filtering.”

In forensic musicology, filtering protectable expression from the unprotectable is one of the core moves of music-copyright analysis. The I-V-vi-IV progression we use is not ours either. A 12-bar blues is not ours. A four-on-the-floor kick pattern, an Alberti bass figure (super familiar thing from Mozart’s Piano Sonata in C Major?), um… ii–V–I jazzy turnarounds, a “boom-boom-clap” beat, none of these were ours before AI came along either. They are scènes à faire: stock figures so common to a genre that protecting them would make for less pleasant music for us all to enjoy. So we apply an abstraction-filtration-comparison process to separate the expressive layer from the foundational layer to clarify what was actually copied and what’s protectable — that’s a lot of what copyright questions are about, the lineage that runs through Arnstein v. Porter and every modern infringement case where the question is whether what was taken was the expression of an idea or merely the idea.

AI-assisted composition doesn’t change that. It extends it. For our purposes here, we’re filtering what a Suno type tool produces by itself, from what was significantly a human’s creative input. So while what it creates might be to some extent “original,” the unprotectable layer of copyright just got thicker. The point though is that the work of identifying what is protectable is pretty much the same work musicologists like Musicologize have been doing since Tin Pan Alley.

What “your creativity” might look like in an AI-assisted workflow

Three scenarios, all using Suno because for now they seem to me to be leading the pack.

You write the lyrics and the topline melody. Suno generates the backing track. Your lyrics are protectable as a literary work, like a poem. Your melody is protectable as a musical work. The Suno-generated chord pad, drum loop, and bass line are not yours. They’re also not Suno’s, really. Again, the copyright office says AI input is not registerable. Register the song though. The registration covers your lyrics and your melody. The backing track is disclaimed, effectively filtered out.

You generate a vocal stem with AI, then rewrite and re-arrange it. The report includes that “if a user edits, adapts, enhances, or modifies AI-generated output in a way that contributes new authorship, the output would be entitled to protection… the copyright would extend to the material the human author contributed but would not extend to the underlying AI-generated content itself.” Let me be careful not to make this any simpler than possible. Your rewrites, your edits, your arrangement choices, your selection of which fragments to keep and which to scrap, that is registrable. An unmodified output from AI would not be. And this is separate from the idea/expression filtering we talked about. I still might do an analysis on your piece and find that it likely infringes on someone else’s work, or that someone else who does basically the same melody as yours likely isn’t infringing on yours!

You might use AI for ideation, experiments, inspiration — a starting point — but then you write the final song yourself. The final song is fully your creation, same as if you used a Beatles song as your inspiration. The copyright office says, “the use of AI as an assistive tool… does not change the copyrightability of the output.” When AI is upstream of the actual writing, it is no different from a chord-suggestion plugin or my Beatles song, or that Mozart. The final fixed expression is yours.

The headline when it came out was “The Copyright Office Won’t Register Works By Artificial Intelligence!” But the two operative phrases are “expressive elements” and “human creative control.” The Office is being extremely reasonable, not a surprise, saying the question is the degree of human control, not the predictability of the outcome. You don’t lose authorship because you can’t perfectly anticipate how the tool’s production will sound. You give up authorship when the expressive choices were the machine’s and you merely accepted them.

How to register

Use the Standard Application. Identify yourself as the human author. In the “Author Created” field, describe your contribution clearly. Don’t just say “song,” but “music and lyrics” or “music, lyrics, and arrangement of vocal performance,” whatever your human creation piece is. In the “Material Excluded” / “Limitation of Claim” fields, disclose the AI-generated contribution that you didn’t significantly alter. Something like, “AI-generated instrumental backing track produced using Suno” is the right level of granularity. Consider the case of Zarya of the Dawn by Kristina Kashtanova. It is an 18-page comic book where the Midjourney-generated images were disclaimed but whose human-written text and human selection-and-arrangement were registered. As of today, no music-specific review board decisions have been issued yet, so visual arts examples are the precedent.

Album registration can save a few bucks.

I started putting this article together when I saw that the copyright registration fee might go up amid some consternation, because many feel independent songwriters are already under attack by AI as a competitor or replacement, along with resentment and lawsuits stemming from the premise that the models were trained on protected material without license or compensation. It’s an insult-to-injury thing. The Group Registration for Works on an Album of Music — GRAM — has been available since March 2021, and it lets you register up to 20 musical works on a single application for one fee. Sorta. There is a separate GRAM application for sound recordings (also up to 20). Both share the same fee. The eligibility rules include that works must share at least one common author across the group, the claimant must be the same for every work, and the works must be first published together as an album.

There’ll be no math. This is just some numbers.

FilingCurrent FeeProposed FeePer-Work, CurrentPer-Work, Proposed
Single-work Standard Application$65$85$65.00$85.00
GRAM, fully loaded (20 works)$65$130$3.25$6.50
GRAM, 10-work mid-load$65$130$6.50$13.00

More pandering to the Copyright Office

If I weren’t a forensic musicologist, I’d be an economist. The Copyright Office last adjusted fees in March 2020. From March 2020 through March 2026, the consumer price index (BLS CPI-U) rose from 258.115 to 330.213, a cumulative increase of nearly 30%. The proposed average fee increase is a bit more at 43%. But they haven’t been raising it along the way, and again, that GRAM thing is pretty darn reasonable even after the proposed increase.

YearStandard Reg. FeeIncrease
1999$20 → $30+50%
2006$30 → $45+50%
2014$35 → $55+57%
2020$55 → $65+18%
2026 (proposed)$65 → $85+31%

Pricing into the Land-Grab.

Let’s look more at the subject of that article in Billboard — In early May 2026, an eleven-organization coalition led by the American Association of Independent Music (A2IM) filed a letter opposing the fee proposal, characterizing the increase as disastrous for independent labels and creators already navigating compressed streaming royalties and stagnant per-stream rates. The framing: a 43% hike with no ability to pass costs through is a barrier to the registration system, not legitimate cost recovery by the office.

The letter is right, and every Musicologize client knows that registration is the predicate for everything. Registering your copyright is optional and the work is “copyrighted” at creation in fixed form, but you can’t protect the work until your register it. And if you don’t register it before someone infringes upon it, things like statutory fees, attorney fees, and other stuff are forfeited (talk to a lawyer; I’m but a musicologist). The framing naturally concentrates on that single-by-single registration as the baseline. As we saw, with GRAM on the table, the per-work cost falls to $3.25 today and $6.50 proposed.

I half-joked on Twitter that it’s rational for the copyright office to lean into higher prices because newly empowered producers can get an album recorded in an afternoon of effective prompting — the time it used to take to get 32 bars into Logic Pro. They’re in no frame of mind to bristle at the registration fee! A fully loaded GRAM registration today comes out to $3.25 per work. Under the proposal, $6.50 per work. So if you can GRAM, that is the actual price of registering a song in 2026. It requires a common author, but that rule is easily met and even encouraged in an AI-assisted solo workflow, since there is often only one human in the chain. I’ll rustle feathers saying this, but why stick to the more traditional collaborative workflow, co-writers, producer(s), contributing musicians on other instruments you don’t play? That workflow has to navigate the common-author rule pretty carefully. The solo creator using Suno breezes by, relatively speaking. Architecturally, AI-assisted solo workflows are easier to register cleanly under GRAM than traditional collaborative ones are. I am more than willing to be convinced otherwise, but that sounds great to me. The AI-assisted solo workflow clears it more cleanly because the common-author rule is satisfied by definition. The fee structure rewards the workflow that requires the least human creative effort. That’s pricing into the land grab — the Office finding the price point at which AI-assisted registrants pay without flinching. Carving out a little bit of the value AI creates.

I don’t know or even really think that’s what they’re doing. Back to best practices…

Keep Your Receipts!!

The same sort of evidence trail that helps in infringement disputes also helps defend an authorship claim. Treat your own catalog similarly to how a forensic musicologist would for analysis. Keep it all.

Voice memos and ideation captures. Save as much of it as you can stand. The metadata is the timestamp, and the timestamp is the priority date. Logic and Pro Tools preserve the creation date in the project bundle. Just make more copies and try to resist the urge to consolidate and rename. Preserve the messy contemporaneous record!

Prompt logs. Every Suno or Udio session generates a prompt history. Export those! Do not rely on the platform to retain it.

Session files with version history. Logic Pro, Pro Tools, Ableton Live, and Reaper all support project versioning. Use it. Save numbered iterations rather than overwriting. The version history is the work product that shows your selection-and-arrangement decisions over time, just the sort of thing the Office is asking for when it says “human creative control over expressive elements.”

Lyric documents. Write lyrics in a system with version history — Google Docs, a git repository, or even Apple Notes with iCloud sync. Drafts matter. The first scribbled couplet that became the chorus is more valuable as authorship evidence than the polished final lyric sheet, because the polished sheet exists for everyone, and the scribbled couplet exists only for you.

Stems and bounces. Keep the AI-generated stems separate from the human-modified stems. If you re-arranged a Suno-generated backing track, save the original Suno output, save your arrangement project file, and save the rendered final. The three together demonstrate the modification.

If this sounds like overhead, it is. But try to acclimate to it. Everything is shifting. That’s just how it is. And speaking as a forensic musicologist, this is much the same as what stands up under examination when someone effectively asks what you actually contributed. The answer is better supported by files than by stories.

Bottom line

The parts you wrote are yours. Register them. Use GRAM if you can take advantage of bulk pricing. Document everything.

FAQ

Q: Can I copyright a song I made with Suno or Udio? A: Yes — the parts you wrote. Your lyrics and your original melody are protectable. The AI-generated backing track, instrumentation, or vocal stem must be disclaimed in the registration. The U.S. Copyright Office has registered hundreds of works that contain AI-generated material under this framework since 2023, and the Supreme Court declined to disturb the human-authorship requirement on March 2, 2026.

Q: How do I register an AI-assisted song with the Copyright Office? A: Use the Standard Application. Identify yourself as the human author, describe your specific human contribution in the “Author Created” field (e.g., “music and lyrics”), and disclose and disclaim the AI-generated material in the “Limitation of Claim” / “Material Excluded” fields. Be specific about which tool produced the AI material.

Q: Is it cheaper to register a whole album at once? A: Yes, and substantially. But you have to satisfy its requirements. Group Registration for Works on an Album of Music (GRAM) covers up to 20 musical works in one filing. At the proposed 2026 fee of $130, that is $6.50 per song if you max it out, compared to $85 per song for a single Standard Application.

Q: What should I keep on file to defend my authorship of an AI-assisted song? A: As much as you can. Don’t be too much of a neatnik. Voice memos with native device metadata intact, exported prompt logs from Suno or Udio, versioned DAW session files, lyric documents with edit history, and the AI-generated stems kept separate from the human-modified stems. Authorship is proven with contemporaneous files, not retrospective stories.

Q: Do I need a forensic musicologist to register an AI-assisted song? A: No. Registration is something you can complete yourself. A forensic musicologist becomes useful when the authorship question is contested, say an infringement claim involving AI-assisted material on either side, an expert opinion needed for litigation or settlement posture, or pre-release clearance on AI-assisted catalog work where the protectable-expression analysis matters for risk. It’s a little surprising to many that while AI can’t copyright its work, it CAN infringe on other creators’ intellectual property.

Brian McBrearty

Brian McBrearty is a forensic musicologist and music copyright expert witness. He provides clearance opinions, expert reports, and expert witness testimony in music copyright matters. His analysis has been cited in the Pepperdine Law Review, on NPR's All Things Considered, and by Reuters, BBC, and Courthouse News. He is the founder of Musicologize.

Brian McBrearty

Brian McBrearty

Forensic Musicologist

Brian McBrearty is a forensic musicologist based in New York. He provides clearance opinions, expert reports, and expert witness testimony in music copyright matters. His analysis has been cited by Reuters, BBC, NPR, and in the Pepperdine Law Review. The initial call is free.

1 thought on “Can I Copyright a Song I Made with AI?”

Comments are closed.