Short answer. Usually yes to use, often no to own, and the residual risk sits upstream in training data rather than in your licence. A paid plan from a major generator typically grants commercial use rights by contract, but the US Copyright Office holds that prompts alone do not make a user the author — so purely AI-generated audio is not protected and anyone can copy it. Meanwhile the major-label litigation has partly resolved into licensing deals, which is the genuinely encouraging development, while some cases remain live.
The most common mistake in this area is treating one question as two answers to the same thing. This piece separates permission from ownership, sets out what you can actually own, locates where the legal risk really sits, and gives the practical steps for commercial use. AI-generated music here means audio produced by a generative model such as Suno or Udio from a text prompt, with no separately recorded human performance.
This article summarises publicly reported developments and is not legal advice. Anyone making commercial decisions should take advice on their own facts and jurisdiction.
Key takeaways
- A paid platform plan can grant commercial-use rights by contract even though the resulting audio may not be copyrightable at all.
- The US Copyright Office's January 2025 Part 2 report states that prompts alone do not give a user sufficient control to be the author of AI-generated output.
- Universal Music Group settled its lawsuit against Udio in October 2025 and Warner Music Group settled with Suno in November 2025, both bundled with licensing deals rather than shutdowns.
- Sony's litigation against Suno and Udio remained live into 2026, and Germany's GEMA won a Munich court ruling against Suno in mid-2026 over training-data infringement.
- None of the current label litigation targets end users of these platforms; the legal exposure sits with the platforms themselves.
Why is "can I use it" a different question from "do I own it"?
One is a contract with a platform and the other is copyright law, and they can point in opposite directions. Platform terms are the contractual licence a generator grants its users; copyright is the separate, statutory right to stop others copying a work.
Reporting on Suno's terms indicates that paid subscribers receive Suno's assigned rights in output created during a Pro or Premier subscription, while free-tier users are limited to personal, non-commercial use. That is a real, useful commercial permission, and it is why most business use of AI audio is straightforward. Reporting on the same terms notes that Suno also states it cannot guarantee copyright will vest in that output — a candid acknowledgement rather than a caveat buried in small print.
So a brand using AI music in an advert is generally permitted to do so, and generally cannot stop a competitor using the identical track. Both are true at once.
What can you actually own?
The human parts of the work, because the Copyright Office's position turns on control, not effort.
The US Copyright Office's Part 2 report on copyright and AI, published in January 2025, stated that prompts alone do not provide sufficient human control to make users of an AI system the authors of the output. Commentary consistently reads this as meaning purely AI-generated audio lacks the human authorship copyright requires.
What survives that test is the human contribution layered around the output:
- Original lyrics you wrote — your creative work and protectable regardless of how the music was generated. Practitioner guidance consistently identifies this as the single strongest step available, similar in spirit to the human review layered into managed AIGC video production, where a person's editorial decisions are what make the finished asset defensible.
- Arrangement and selection decisions — where you cut, sequence, edit and combine generated material into something whose expressive shape is yours.
- Recorded human performance added to or replacing generated elements.
- The finished hybrid work, registrable on the basis of the human authorship it contains, with the AI-generated material disclosed.
One practical consequence: if ownership matters, record the human work — session histories, drafts, versions, who did what and when. Producers keep timestamped session documentation precisely to evidence meaningful authorship, the same discipline behind human-in-the-loop review of AI content, where the human contribution is recorded as work happens rather than asserted afterwards.
Where does the real legal risk sit?
Upstream, in what the models were trained on, and it has not fully resolved.
The RIAA filed copyright suits against Suno and Udio in June 2024. Since then the picture has split several ways:
| Development | Status |
|---|---|
| Universal Music Group v. Udio | Settled October 2025, reported with licensing arrangements and artist opt-in provisions |
| Warner Music Group v. Suno | Settled November 2025, bundled with a licensing deal for new models launching in 2026 |
| Sony Music v. Suno / Udio | Remained live through mid-2026, with litigation continuing on separate tracks |
| GEMA v. Suno (Germany) | Munich Regional Court ruled for GEMA in mid-2026 on specified uses of protected compositions, covering training and outputs, with disclosure and damages ordered |
Two points matter for a commercial user. First, none of this litigation targets end users; the exposure sits with the platforms. Second, outcomes could change platform terms, catalogues or availability, which is a continuity risk rather than an infringement risk for you. Businesses weighing whether to build in-house versus outsource AI video and audio production at scale often treat this platform-level uncertainty as one of the deciding factors.
What is the encouraging news, and how do you use AI audio safely?
The settlements are the good news, and they point somewhere better than litigation was heading.
Licensing has been beating prohibition. The labels moved from trying to shut these platforms down to signing deals with them: reporting on the Universal-Udio settlement describes a licensed platform launching in 2026 using authorised catalogue as training data, with revenue sharing back to rights holders. That is healthier than either side winning outright — creators get legitimate tools, rights holders get paid. The Warner-Suno deal was reported in similar terms, with artist opt-in provisions addressing the consent objection that drove much of the original anger. With major labels as commercial partners rather than plaintiffs, a wholesale purge of AI-assisted music from distribution platforms becomes far less likely, and nothing in the Copyright Office position penalises AI assistance — it requires human authorship, which most real production work has anyway.
Practical steps for commercial use:
- Use a paid tier and read its grant. Free tiers commonly restrict to personal, non-commercial use, and that is where most accidental breach of platform terms happens.
- Prefer platforms with licensed training data for high-value or long-lived work, since that is where upstream risk concentrates.
- Add and document human authorship if you need to own the result rather than merely use it — the same governance model that underpins disclosure and provenance in AI content.
- Avoid artist imitation. Prompting for a named artist's voice or style invites right-of-publicity and passing-off problems outside the copyright question altogether.
- Match the licence to the use. Background audio for an internal video is a different risk profile from a national campaign or a brand sonic identity, a distinction covered further in synthetic voiceover quality and loudness standards.
- Keep records of tool, plan, date, terms version and what a human contributed, the same discipline outlined in building a licensed voice library for synthetic speech.