Suno v6 Is Trained on Licensed Music. Does That Make Your Song Copyrightable?

Suno v6 dropped on September 9, 2026, and within hours my inbox filled with the same question phrased ten different ways. If the new models were built on music Warner, BMG and Believe actually licensed, does that finally mean the songs coming out of Suno are legally mine, copyright and all?

The short answer is no. A licensed training set changes what Suno can be sued for. It does not change the rule that decides whether you can own a copyright in the output, and that rule, in the United States, is about human authorship. Nothing about Suno v6 copyright has moved on that front. Licensed data lowers one specific category of risk for Suno and, indirectly, for you. It does not hand you an enforceable copyright, and it does not make every v6 track clear for every use.

I make music with these tools every week, I have registered AI-assisted songs with the Copyright Office, and I have read Suno’s terms more times than any healthy person should. So let me walk through what v6 changed, what it did not, and what you can do about the gap.

What Actually Changed With v6

The v6 family is three models. There is the flagship v6, which Suno describes as its reliable and precise option, a more experimental v6-wild for people who want unpredictable results, and a faster v6-mini that free accounts can use. Suno’s own launch post says the new generation was developed with industry partners including Warner Music Group, BMG and Believe, and that older models are being retired as v6 rolls out. According to the TechCrunch report on the v6 launch, Suno says v6 was not trained on the data behind its previous models.

That last sentence is the real news. Suno is not saying v5.5 got a licensing patch. It is saying the old models, the ones at the center of the label lawsuits, are being pulled from the product and replaced with something built from a different dataset. Music Ally’s coverage frames it as an attempt to draw a clean legal line between old Suno and new Suno, while noting that the litigation over the old models does not go away because those models are gone.

Two details matter for the copyright question. First, the training data is not purely licensed catalog. Suno says it also includes “user data,” meaning songs people generated during the era of the unlicensed models. Second, as the Lapaas Voice analysis of the licensed models points out, licensed does not mean every recording in the world. It means specific partner catalogs and opt-in arrangements, and the exact scope of repertoire and territories has not been published in full.

Pricing did not move with the launch. The free plan gives you 50 credits a day on v6-mini, no downloads and no commercial rights. Pro is $10 a month, or $8 a month billed annually, and includes 2,500 credits, access to v6 and v6-wild, commercial use rights and 20 song downloads a month. Premier is $30 a month, or $24 billed annually, with 10,000 credits, 60 downloads a month and Suno Studio. Those figures are from Suno’s own pricing page, and the download caps matter more than they look.

Two Different Questions That Keep Getting Mashed Together

Almost every confused take I have seen this week collapses two separate legal questions into one.

Question one is about the input. Was it lawful to train the model on the music it learned from? That is what UMG and Sony are fighting Suno about in Massachusetts, and what GEMA fought Suno about in Munich. Licensing deals with Warner, BMG and Believe address this question, at least for those catalogs.

Question two is about the output. When the model generates a song, is there a copyright in it, and who holds it? That is governed by the US Copyright Act, and it does not care where the training data came from. A model trained on flawlessly licensed music and one trained on pirated music produce outputs that are treated identically for authorship purposes. What decides copyright is who made the creative choices, not what the model listened to.

Once you separate those two questions, the v6 announcement makes sense. Suno is fixing a training problem. It did not, and could not, fix an authorship problem, because that one is not Suno’s to fix.

The Human Authorship Rule Has Not Moved

Here is where the law actually sits as of September 2026.

The US Copyright Office published Part 2 of its report on copyright and artificial intelligence in January 2025, and it is still the operative guidance. The report’s conclusion is blunt: given the technology generally available, prompts alone do not give a user enough control over the output to make that user the author. The Office describes prompts as instructions that convey ideas, and ideas are not protectable. The Copyright Office’s copyrightability report is worth reading in full, because most summaries get the nuance wrong.

The Office did not say AI-assisted works are unprotectable. It said three things can support a copyright claim: when human-authored material is perceptible in the output, when a human uses AI as a tool to expand on their own protectable input, and when a human makes creative selections, arrangements or modifications of the output. It also noted that it had already registered hundreds of works incorporating AI-generated material, with the AI portions disclaimed.

The courts have backed the Office. In March 2025 the D.C. Circuit decided Thaler v. Perlmutter, holding that the Copyright Act requires a work to be authored in the first instance by a human being. Then, as Mayer Brown’s note on the Supreme Court’s decision explains, the Supreme Court declined to review that ruling on March 2, 2026. That closed the door on the argument that a machine can be an author.

What Thaler did not decide is how much human involvement is enough when a person uses AI as a tool. The Stark & Stark analysis of ownership challenges points to Allen v. Perlmutter in Colorado, where the question is whether iterative prompting plus post-generation editing adds up to authorship, with summary judgment motions still pending as of August. So the edges are being litigated. The center is not. Typing a style prompt and clicking generate five times until you like the result is not authorship under current US law, and no licensing deal changes that.

The one sentence version: the training license answers who Suno owes, not who owns the song.

What Suno’s Own Terms Actually Give You

People say “Suno gives you full ownership on Pro.” That is half true, and the other half is the important half.

Suno’s terms of service, in the version that took effect September 3, 2026, say that for Pro and Premier subscribers, Suno assigns to you all of its right, title and interest in any output generated from your submissions. Read that carefully. Suno is assigning whatever it has. It is not creating a copyright, because it cannot. In the very next sentence the terms state that because of the nature of machine learning, Suno gives no warranty that any copyright will vest in any output. Suno is telling you, in its own document, that the thing it just assigned to you might be empty.

Suno’s help center puts the free tier plainly: on the Basic tier, Suno retains ownership of what you generate, and you may only use it for non-commercial purposes. And the current Suno terms of service add a few restrictions that even paid users tend to miss.

The commercial use right only attaches to output you have downloaded through an approved channel, within your monthly allotment. That is 20 downloads on Pro and 60 on Premier, with extra downloads available for purchase. A track sitting in your library that was never downloaded, or one you captured by recording the stream, cannot be commercially exploited under the contract. The terms also let Suno append fingerprints, watermarks or metadata showing tier and download status, and you agree not to strip them.

Remixes of other users’ songs are joint works that may only be used non-commercially regardless of tier. Outputs are not guaranteed to be unique, and someone else can receive the same or similar output from a similar prompt. And a new line says applicable rights holders may have the right to collect revenue from distribution of outputs on third-party platforms. That is the licensing deals showing up in your contract. Music Ally reported that revenue sharing with Warner and BMG starts with the v6 rollout, so some slice of what your v6 track earns is, in effect, spoken for.

So a paid plan gives you a strong contractual right to use your downloads commercially, which is real and valuable. It gives you a copyright only to the extent one exists, and for a prompt-only track, one usually does not.

The Risks a Licensed Pipeline Does Not Remove

To be fair to Suno, licensed training data is a large improvement, and I would rather generate on a model where the labels are paid than on one where they are not. But several things are still hanging over the tool, and they affect whether your track is clear for the uses you have in mind.

The US label litigation is still live. Warner settled with Suno in November 2025, but UMG and Sony did not. As the Chartlex litigation tracker summarizes, no US court has ruled on whether Suno’s training was fair use, and court schedules now push any such ruling into 2027. Music Business Worldwide reported that UMG and Sony moved to expand the case to more than 61,000 recordings, which at the statutory maximum of $150,000 per work would be a theoretical exposure north of $9 billion. That case is about the old models, but the company you are relying on for your download rights is the same company.

Then there is Munich. On July 31, 2026, the Munich Regional Court ruled largely for GEMA, the German collecting society, against Suno. According to Bird & Bird’s summary of the GEMA v Suno judgment, the court found infringement not only in the US training but also in the outputs, because Suno generated songs substantially similar to compositions like “Daddy Cool” and “Rasputin.” Music Week reported that Suno must disclose the scale of its use and pay damages still to be determined. Suno has said it is evaluating an appeal.

Why does a German case about old models matter to a US musician using v6? Because it shows the output side has its own risk. A model can memorize and regurgitate. Licensed training reduces the odds that a partner’s melody ends up in your chorus without permission. It does nothing for melodies belonging to UMG, Sony, an independent label, or the indie artists who have filed their own suits. AI Vortex’s case summary notes a proposed class action by independent musicians, Nguyen v. Suno, filed in the Northern District of California in November 2025, arguing that the major-label deals only cover major-label catalogs. And remember that “user data” line in the v6 training description. Songs generated during the unlicensed era went back into the new model.

The practical lesson is that “licensed” describes a contract between Suno and specific partners. It is not a warranty that your specific output is free of anyone else’s protectable expression, and Suno’s terms disclaim exactly that. If you ever get a claim on a v6 track, “the model was licensed” is a better position than you had a year ago, but it is not a defense you can wave at a court and walk away.

What Distributors and Spotify Actually Care About

The copyright question and the distribution question are also different, and the second one is where most people hit real friction.

Spotify’s September 2025 policy overhaul is still the frame. The company said it had removed more than 75 million spam tracks in the preceding year, launched a spam filter aimed at mass uploaders, banned unauthorized voice clones, and backed a DDEX metadata standard for disclosing where AI was used in a track. DDEX is simply the industry’s shared data format for delivering music and credits to streaming services. As Music Ally’s coverage of Spotify’s AI measures listed, Believe, CD Baby and DistroKid were among the first distributors to sign on to the disclosure standard. Spotify’s line is that disclosure does not get you penalized, but non-disclosure of something later flagged as AI can.

Distributors sit in between, and they do not agree with each other. In my experience DistroKid is the most workable option for AI-assisted releases as long as you tick its AI disclosure box and actually own or control every element. CD Baby has been the strictest, with multiple 2026 guides describing it refusing fully AI-generated music. TuneCore has said it will not take 100 percent AI tracks and, as the MUSICΔ comparison of distributor terms reported in early 2026, raw Suno output submitted unedited was frequently rejected there.

Then v6 changed one thing. Music Ally reported that Believe, which owns TuneCore, had banned Suno-made tracks from its pipeline in April 2026, and that the v6 announcement says tracks made with the new industry partner model will become eligible for distribution through Believe and TuneCore. That is a direct consequence of licensed training data, and a real benefit. But notice what it is. A distributor deciding to accept your file is not a court deciding that you own it. Getting on Spotify and holding a copyright are two different achievements, and you can have either without the other.

Whatever distributor you use, disclose. Fill in the AI fields accurately, list which parts were generated, and keep your Suno receipt and download record for each track. Undisclosed AI content that later gets flagged is the fastest way I know to lose a release and the streams attached to it.

Where the Line Actually Sits: What Makes a Suno Track Yours

So what does put a copyright in your hands? The same things that always did. Human creative expression that you can point to.

Lyrics are the easiest. If you wrote the words yourself and pasted them into Suno, they are a literary work you authored, regardless of how the music was made. If Suno wrote the lyrics from your prompt, they are not yours.

Performance is next. A vocal you sing, a guitar part you record, a bass line you play over the generated track: each is a human performance fixed in a sound recording, and that recording is protectable. Suno has quietly leaned into this. Pro and Premier let you record your own voice and build a voice model from it. If your actual singing is the lead vocal, your authorship claim is much stronger than if the model’s vocal is.

Arrangement and structure count when they are yours. The Copyright Office’s guidance cites Zarya of the Dawn, a comic where the author’s text and the selection and arrangement of the AI images were registered while the images themselves were not. The RightsDocket guide to registering AI-assisted music draws the practical lesson for musicians: do not frame your claim around hours spent prompting. Frame it around the lyrics you wrote, the structure you built, the performances you recorded and the substantive edits you made. If you took v6 stems into a DAW, cut them, re-sequenced them, layered your own parts and shaped the final mix, you have a selection and arrangement argument. If you exported the two minute thirty second output and uploaded it, you do not.

The v6 editing tools are a trap here. Telling the model “change the chorus to a gospel choir” or swapping one lyric is convenient, and it makes the output feel more like yours. Legally, it is still a prompt. You describe what you want and the model decides how to express it. Do not confuse having more knobs with having authorship.

Registration is where this gets tested. You file with the Copyright Office, describe the human-authored material, and exclude the AI-generated material in the limitation of claim section. I have done this and it is not scary. The examiner wants a clear, truthful account of what you made. The registrations that get refused are the ones that claim the whole thing.

There is also the blunt option. When a Suno track has become something you care about and want to own outright, the cleanest route is to have real musicians re-record it, using the generated version as a demo. The composition still has an authorship question if the melody came from the model, so a human writer should rework the melodic material rather than copy it note for note, but the master recording becomes fully human-made. It costs more than a subscription, and it is the only path I know that produces a track with no asterisks on it.

My Workflow for a v6 Song I Plan to Release

Here is what I actually do now, in order.

I write the lyrics first, in a text file with a timestamp, before I open Suno. That single habit gives me a dated, human-authored literary work for every song.

I generate on Pro or Premier, never on the free plan, because free tier output is non-commercial and upgrading later does not retroactively fix songs you made while free. I use the flagship v6 for anything I intend to keep and v6-wild only for idea hunting.

I download every keeper through the app, not by any other route, and I note the download in a spreadsheet with the date, the plan I was on and the prompt. Those 20 or 60 monthly downloads are the thing that turns a library item into a commercially usable file under the terms, so I treat them as the scarce resource they are.

I pull stems, either with Suno’s stem separation or in Suno Studio on Premier, and bring them into my DAW. Then I do real work. I replace the lead vocal with my own take when the song matters, re-record at least one instrumental part live, rearrange sections, and mix and master the result myself. I am not doing this to fool a detector. I am doing it because that is the work that creates authorship.

Before release I check the track for resemblance to existing songs. Nothing is bulletproof, but if a melody rings a bell, I change it. The Munich judgment is a reminder that similarity, not intent, is what a court looks at.

I disclose AI use on the distributor’s form, describe exactly which layers were generated, and keep the Suno receipt, download log and DAW session together in the release folder.

Then I register with the Copyright Office, claiming the lyrics, my performances and the arrangement, and excluding the generated material. Check the current fee schedule on the Copyright Office site before filing, because fees change.

If the song is one I want to pitch for sync, I skip most of the above and get it re-recorded properly. Sync supervisors ask for chain of title, and “assigned by Suno, copyright not warranted” does not close deals.

A Reality Check, Including for UK and EU Readers

Everything above is written from the US frame, which is where Suno is based and where its terms are governed. The rules differ elsewhere, and it is worth being precise about how.

In the UK, section 9, subsection 3, of the Copyright, Designs and Patents Act 1988 has long said that for a computer-generated work with no human author, the author is whoever made the arrangements necessary for its creation. On its face that could cover a Suno user. But as Herbert Smith Freehills Kramer’s summary of the March 2026 government report notes, the UK government’s stated preference is to repeal that provision and protect only AI-assisted works where a human made creative choices. No bill had been introduced as of that report, so the provision technically still stands, but I would not build a release strategy on a rule the government has said it wants gone.

In the EU, copyright arises from the author’s own intellectual creation, and the AI Act adds transparency duties for model providers rather than a new ownership right for users. The Munich ruling shows EU courts will hold a US model liable for outputs served to EU users, which means more scrutiny, not less. Wherever you release from, a prompt-only track is on shaky ground and a human-assisted one is defensible.

There are also things nobody outside Suno and its partners knows yet. The full terms of the Warner, BMG and Believe deals have not been published, the opt-in artist products Suno has teased have no dates or prices, and v6 is two days old as I write this, so nobody has enough testing to say how often it reproduces recognizable material. Treat any confident claim on those points as a guess.

One Thing to Do This Week

Go into your Suno library and look at the tracks you have released or plan to. For each one, write down, in one line, what a human actually authored. Lyrics you wrote, a part you played, a structure you built, or nothing.

The ones where the answer is nothing are the ones licensed training data did not save. Pick your favorite, open a DAW, and put something of yourself into it. That is the entire difference between a song Suno let you use and a song that is yours.

Sources

• Suno, Introducing v6: https://suno.com/blog/introducing-v6
• Suno, Terms of Service: https://suno.com/terms-of-service
• Suno, Pricing: https://suno.com/pricing
• Suno Help Center, Does Suno own the music I make?: https://help.suno.com/en/articles/2416769
• TechCrunch, Suno replaces its AI models with a new one trained on licensed music as copyright suits pile up: https://techcrunch.com/2026/09/09/suno-replaces-its-ai-models-with-a-new-one-trained-on-licensed-music-as-copyright-suits-pile-up/
• Music Ally, Suno launches its v6 AI-music models. Here’s what you need to know: https://musically.com/2026/09/09/suno-launches-its-v6-ai-music-models-heres-what-you-need-to-know/
• Lapaas Voice, Suno v6 Rebuilds AI Music Around Licensed Models: https://lapaasvoice.com/suno-v6-licensed-ai-music-models
• U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability: https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf
• U.S. Court of Appeals for the D.C. Circuit, Thaler v. Perlmutter: https://media.cadc.uscourts.gov/opinions/docs/2025/03/23-5233.pdf
• Mayer Brown, Supreme Court Denies Cert in AI Authorship Case: https://www.mayerbrown.com/en/insights/publications/2026/03/supreme-court-denies-review-in-ai-authorship-case
• Stark & Stark, OUTPUT v. INPUT: Copyright Ownership Challenges in the Era of Artificial Intelligence: https://www.stark-stark.com/news/output-v-input-copyright-ownership-challenges-in-the-era-of-artificial-intelligence/
• RightsDocket, The Complete Guide to Registering AI-Assisted Music with the U.S. Copyright Office: https://www.rightsdocket.com/insights/register-ai-music-usco-guide
• Chartlex, Music Industry AI Lawsuits Tracker 2026: Live Status: https://www.chartlex.com/blog/business/music-industry-ai-lawsuits-tracker-2026
• Music Business Worldwide, After a judge denied Sony’s move to expand its Udio case, Suno asks court to reject UMG and Sony’s bid to add 61K recordings: https://www.musicbusinessworldwide.com/after-a-judge-denied-sonys-move-to-expand-its-udio-case-suno-asks-court-to-reject-umg-and-sonys-bid-to-add-61k-recordings/
• AI Vortex, Suno and Udio AI Music Lawsuit: 2026 Status & Timeline: https://www.aivortex.io/legal/ai-case-law/suno-udio-music-ai/
• Bird & Bird, Munich District Court Rules on AI-generated music GEMA v Suno: https://www.twobirds.com/en/insights/2026/germany/munich-district-court-rules-on-ai-generated-music-gema-v-suno
• Music Week, GEMA wins court ruling on breach of copyright by AI music firm Suno: https://www.musicweek.com/publishing/read/gema-wins-court-ruling-on-breach-of-copyright-by-ai-music-firm-suno/094644
• Music Business Worldwide, Spotify has deleted 75m+ tracks in ‘spammy’ AI music crackdown: https://www.musicbusinessworldwide.com/spotify-has-deleted-75m-spammy-tracks-as-it-unveils-new-ai-music-policies/
• Music Ally, Spotify reveals its latest measures to handle AI music, spam and deepfakes: https://musically.com/2025/09/25/spotify-reveals-its-latest-measures-to-handle-ai-music/
• MUSICΔ, AI Music Distribution Terms Compared: DistroKid, TuneCore, and CD Baby: https://musicdelta.com/en/articles/ai-music-distribution-terms
• Herbert Smith Freehills Kramer, UK Government Report on Copyright and AI concludes more evidence is needed: https://www.hsfkramer.com/notes/ip/2026-03/uk-government-report-on-copyright-and-ai-concludes-more-evidence-is-needed-although-s9-3-cdpa-could-go