If you make music in the UK with Suno, Udio or any other generative tool, you may have read that British law is unusually generous to AI: section 9(3) of the Copyright, Designs and Patents Act 1988 (CDPA) gives an author to "computer-generated" works. It's tempting to treat that as your safety net. We wouldn't. The government has said it wants to remove that protection, and for recordings it was never the right section anyway. This guide explains why, and sets out a practical plan for protecting the recordings you make with AI help using the parts of the law that look set to stay.
A quick note before we start: we're one human songwriter and Petunia, our AI collaborator, not lawyers, and this isn't legal advice. If a release matters commercially, or you're signing a deal, talk to a UK music or IP lawyer. If you're registering songs in the United States, the rules are different; our guide to building a human-authorship paper trail covers that.
Why Section 9(3) Is a Shaky Foundation
Section 9(3) says that for a literary, dramatic, musical or artistic work that is computer-generated, the author is "the person by whom the arrangements necessary for the creation of the work are undertaken". Section 178 defines computer-generated as made in circumstances where there is no human author. It has been in the Act since 1988, and the UK is one of very few countries with anything like it.
The problem is that it sits awkwardly with how originality now works. To be protected, a song or lyric (a musical or literary work) has to be original, and the courts now treat that as meaning the author's own intellectual creation, reflecting free and creative choices. A work with no human author is hard to square with a test built around a human's creative choices. Commentators, including Clifford Chance back in 2023, flagged that tension, and also pointed out that it's unclear whether the person typing prompts or the company running the tool would count as the person making the "arrangements".
On 18 March 2026 the government published its Report on Copyright and Artificial Intelligence, required under the Data (Use and Access) Act 2025. On computer-generated works it says most consultation responses wanted works created solely by AI left unprotected, and it proposes that this "specific type of protection should be removed", while copyright continues to protect works created with AI assistance. It also says it will keep monitoring first and won't change the law until it's confident, so there's no date yet. As Mishcon de Reya put it, protection looks likely to go eventually, but the decision has been deferred. Either way, building your ownership on s.9(3) means building on something the government has said it plans to take away.
The Good News: Recordings Work Differently
A finished track usually carries at least two separate copyrights. The musical work (melody and arrangement) and the lyrics (a literary work) need originality. The sound recording itself is protected under section 5A, and it has no originality test at all. Its author, under section 9(2)(aa), is the producer, which section 178 defines as the person by whom the arrangements necessary for making the recording are undertaken.
That matters because s.9(3) only covers literary, dramatic, musical and artistic works. Sound recordings are what lawyers call entrepreneurial works, alongside films and broadcasts, and they've never depended on it. Mishcon de Reya's summary of the report notes that sound recordings stay protected even if generated by AI, because they have no originality requirement. Two limits are worth knowing. Copyright doesn't exist in a recording to the extent it's a copy of a previous recording (s.5A(2)). And the term is 50 years from the end of the year it was made, or 70 years from the end of the year it was first published or made available (s.13A).
So the practical plan is: claim the lyrics and music you actually wrote as yours, treat the recording as protected through the producer route, and use evidence, contracts and metadata to back both up. One caution: whether the user, the platform or someone else is the "producer" of a recording generated inside an AI app hasn't been tested in court. Your tool's terms of service may also say who owns or licenses the output, so read them for the plan you're on.
Step by Step: Protect What You Make
Step 1 — Read your tool's terms before you release. Look for who owns the output, what licence you get, whether commercial use needs a paid plan, and whether that changes if you downgrade. Save a dated PDF or screenshot of the terms in your project folder, because terms change.
Step 2 — Write down the "arrangements" you made. For each recording, note who chose and paid for the tool, who set up the session, who supplied the lyrics and style direction, who chose the takes, and who exported, edited and mixed the final file. A one-line entry per session is enough, for example 2026-11-01 · producer: [name] · Suno session on [name]'s paid account · lyrics_v3 + style prompt "late-night soul, warm Rhodes, female lead" · kept take 4. This is the evidence for who undertook the arrangements for making the recording.
Step 3 — Keep your human-written parts separate and dated. Save lyric drafts as numbered files, keep voice memos of melodies you hummed, chord charts and any parts you played or sang yourself. These support the originality of the musical and literary works without any help from s.9(3).
Step 4 — Make your edits recordable. Split stems, re-sing a line, replace a part, rearrange sections, mix and master, and save the DAW project with the date in its name. Real production work strengthens your position as producer and adds human creative choices to the music itself.
Step 5 — Put it in writing with collaborators. If anyone else is involved (a co-writer, a vocalist, a mix engineer, a label), agree in a short written document who owns the recording, who owns the composition and lyrics, and how income is split. A written assignment or agreement settles questions the law hasn't. Something as simple as Recording owned by [name] as producer. Lyrics by [name]. Composition split 50/50 [name]/[name]., signed and dated, is far better than nothing.
Step 6 — Fill in the metadata consistently. When you upload through your distributor, enter the rights owner of the recording (the ℗ line) and the songwriters accurately, and answer any AI-use questions honestly. Use the same names in your PRS for Music and PPL registrations if you're a member, and in your project notes. Mismatched names across systems are a common source of disputes.
Step 7 — Be honest about what AI made. Don't claim to have written lyrics or a melody that the tool produced. Overclaiming weakens the parts you can genuinely defend, and the report shows labelling of AI output is on the government's list: it set up work on labelling best practice, with an interim report promised for the autumn.
Step 8 — Back up and leave it alone. Once a track is out, copy the whole project folder somewhere safe and don't rename files. The dates are part of the record.
Common Mistakes
Relying on s.9(3) for ownership: it may be removed, and it doesn't cover recordings in any case.
Assuming you own whatever the app exports: the platform's terms may decide that. Check and save them.
No paperwork with collaborators: a handshake won't help if a track takes off. Write it down.
Claiming AI-written lyrics or melodies as your own: claim what you made and say what the tool made.
Different names in different places: keep the distributor, PRS, PPL and your notes consistent.
If you're also handling platform AI labels, our guide to documenting an AI-assisted track for platform disclosure uses the same project folder.
Where That Leaves Us
UK law on AI and copyright is still moving, and the government has said it won't rush. The safest approach doesn't depend on how that ends: protect the parts you wrote through originality, protect the recording through the producer route, and back both with dated records, written agreements and consistent metadata. If you'd like to hear the songs we make together, a human songwriter and Petunia, our AI collaborator, they're on the homepage, and our Suno Guide has the prompts, style tags and how-to guides we keep coming back to.
Sources
GOV.UK — Report on Copyright and Artificial Intelligence (DSIT, DCMS and IPO, 18 March 2026; proposal to remove protection for wholly computer-generated works, monitoring first)
Mishcon de Reya — Copyright and AI: The UK Government's report (March 2026; CGW protection likely to go, sound recordings protected without originality, labelling interim report)
Fieldfisher — UK government maintains status quo on AI and copyright (March 2026; no immediate legislative changes, labelling taskforce)
Society for Computers and Law — Keeping Ownership Authentic: AI and Control in the Modern Matrix of Music (June 2026; contracts and documentation as the practical answer)
Clifford Chance — AI-Generated Music and Copyright (April 2023; originality tension, who makes the "arrangements")
legislation.gov.uk — CDPA 1988 sections 9, 5A, 13A and 178 (authorship, sound recordings, term, definitions of "producer" and "computer-generated")
