AI music and copyright: who owns it?

You type a prompt into Suno or Udio, hit “generate” and a dozen seconds later you have a finished track: a verse, a chorus, a mix, sometimes even vocals in your favourite style. The question that follows is always the same: who actually owns this? Can you sell it, put it on Spotify, use it in a client's ad? And if someone copies your track — do you even have anything to chase them with? This piece sorts out what is known today and what is not.

This is general information, not legal advice. Copyright law for AI-generated content is new, unsettled and varies by country. If you are building a real business on AI music, take your specific situation to an intellectual-property lawyer. Below I explain the mechanisms; I do not rule on your case.

The core question: can “pure” AI output be copyrighted?

Counter-intuitively, it is not about “who clicked generate”. It is about a much older question: was a copyrightable work created at all? Copyright in most systems protects human creativity — original expression originating from a person. Traditionally that assumption was so obvious it rarely needed spelling out. Generative music puts it on a knife's edge.

The line emerging across various jurisdictions looks roughly like this: output that issolely the product of a machine — with no meaningful creative contribution from a person — tends to be treated as something that cannot be covered by classic copyright, or at the very least as something surrounded by serious uncertainty. The more human input there is in the process (selection, arrangement, editing, written lyrics, reworking), the stronger the argument that a protectable contribution exists. This is not a hard numeric threshold — it is a spectrum, and courts and offices are only just drawing its edges.

The human-authorship principle, in general terms

“Human authorship” is a concept that runs through the AI debate worldwide, though it takes different forms. The point is simple: copyright exists to protect human expression, not the mere fact that something exists. Several practical consequences follow, regardless of country.

• Merely typing a prompt is often treated as too little — just as commissioning someone to “paint me a sunset” does not make you the author of the painting.
• Contribution that shapes the final track (your arrangement decisions, editing, stitching parts together, your own lyrics, overdubs) builds the argument for protecting the part a human genuinely created.
• What is protected is usually your contribution, not necessarily the entire file the model produced. The boundary is often blurry.

To be clear: the details — how much input is enough, how to prove it, whether a given country requires registration — differ and are sometimes contested. There is no single global answer.

A platform's terms of service are not the same as statutory copyright

This is where most of the confusion starts. When Suno or Udio say you have “commercial rights” to your tracks, that is a contract between you and the platform — a contractual licence. It is not an official confirmation that copyright exists in that track in the statutory sense.

Separate two entirely different layers:

1. The contract layer (platform terms): what the provider lets you do with your output. This usually depends on your plan — a free tier often limits you to non-commercial use, while paid plans typically “unlock” commercial use. Terms can change, so read the current version.
2. The statutory layer (copyright law): whether copyright arose in the work at all, who the rightsholder is, and whether you can stop others from copying it. That is decided by the law of the relevant country, not by the terms.

The practical takeaway: a platform may let you sell a track while at the same time nobody— not you, not the platform — holds a strong, exclusive copyright to stop someone who copies it. “I am allowed to use it” and “I have exclusivity” are two different things.

Commercial-use risks

Suppose you want to use AI music in a product: a backing track for an ad, a podcast intro, a game soundtrack, a track for sale. Here is what to genuinely watch for.

Weak exclusivity

If copyright over the purely machine-made parts is doubtful, your ability to block copying may be weaker than it looks. For a track meant to be a signature of your brand, that is a real risk.

Shifting licence terms

Your right to commercial use flows from the terms, and those can change, and your plan can lapse. Record the plan and conditions under which the track was made, and keep the evidence.

Unintended similarity

Models are often asked for the style “in the style of” a famous artist. If the output is too similar to a specific protected recording, the risk does not vanish just because AI made it. Avoid prompts aimed at a specific track or the voice of a living performer.

Disclosure rules and platform policies

Some distribution platforms, libraries or clients have their own rules for AI content (labelling, limits, bans). This layer is independent of copyright, and it can block your release. Check the recipient's terms, not just the generator's.

Training-data disputes — at a high level

A separate, heated front is the question of what the models were trained on. Music generators learn from vast catalogues of recordings, and disputes swirl around this: whether using protected works for training requires consent and payment, or falls within exceptions. Different countries approach it differently, and many questions are still being decided.

For you as a creator, two conclusions matter. First: this dispute mainly concerns the relationship between data creators and model providers, but risk is sometimes pushed downstream to the user — read the liability and indemnity clauses in the terms. Second: the landscape can shift retroactively once rulings land. What is “allowed” today may require extra conditions tomorrow. Do not treat today's state as permanent.

Polish and EU context, in general terms

In Poland, as across the EU, copyright traditionally ties protection to human creativity — to a “manifestation of creative activity of an individual character”, as it is often described. Purely machine-made output, with no creative input from a person, fits poorly into that assumption. That is a general observation about the direction of thinking, not a quote from a specific provision and not a verdict on your case.

At the EU level there is ongoing work and debate around AI, including the transparency of training data and the labelling of generated content. The direction is towards moredisclosure obligations, not fewer. The specific requirements are crystallising, so any firm claim is worth verifying against the current state and with a lawyer. I am not inventing article numbers or rulings here, because the subject is in motion.

A practical checklist for creators who want to earn from AI music

1. Read the current platform terms for your plan — what is allowed commercially, what the limits are, whether rights lapse after you cancel the subscription.
2. Add real human input. Arrange, edit, write lyrics, combine versions, rework. The more of your creative decisions, the stronger your position over your part.
3. Document the process. Save prompts, versions, dates, the plan and your own edits. Proof of contribution can be priceless when someone asks “who made this”.
4. Avoid imitating specific recordings and voices. “Energetic 80s synth-pop” is safer than “sounds like that one specific hit”.
5. Read the liability clauses. Check who carries the risk of training-data claims and whether the provider offers any safeguard.
6. Check the recipient's rules. The distribution platform, music library or client may require AI labelling or ban such content outright.
7. Do not promise exclusivity you may not have. In client contracts, describe what they actually get, and do not guarantee protection the law may not provide.
8. When real money is involved — see a lawyer. A campaign, a licence, a catalogue sale: a short consultation is cheaper than a dispute.

TL;DR

Purely machine-made musical output — with no creative human input — can be hard to cover with classic copyright, and the legal position is unsettled and varies by country. A platform's terms (Suno, Udio) give you commercial rights at the contract level, depending on your plan, but that is not the same as statutory exclusivity. The more genuine input you add, the stronger your position. Watch out for weak exclusivity, shifting licences, similarity to specific recordings, training-data disputes and recipient rules. Document the process, read the clauses, and when real money is involved, ask a lawyer. This is general information, not legal advice.