AI7 October 20269 min read

Proving your music is human-made: AI disclosure and what to keep

AI music disclosure for composers: what supervisors ask, what the EU AI Act and US Copyright Office actually say, and the files to keep as proof.

By Daniel Angelucci, composer and founder of DADAAUDIO

Supervisors and music buyers increasingly ask whether a track involved AI, and some will only license music made by people. No law makes a composer who doesn't use AI prove anything, but the simplest protection is a habit: keep the working files that show how the music was made, and answer the AI question plainly and in writing. If you did use AI tools, say where and how; societies and copyright offices now have rules for that too.

This article covers why the question is coming up, what the rules really say (and don't), and exactly what to keep. It isn't legal advice: if a deal depends on it, talk to a lawyer or your society.

Why supervisors are asking now

In December 2025, Cyanite, MediaTracks and Marmoset published a survey of 144 music supervisors, filmmakers, advertisers and editors. The headline numbers:

Finding	Share of respondents
Want transparency about whether music is AI-generated or human-made	97 %
Will only license human-made music, regardless of labelling	49 %
Want richer metadata about tracks (cultural background, creative intent)	52 %
Value the artist's cultural background as a selection factor	40 %

It's a small sample, and the same survey found advertising people more open to AI music than film and supervision people. But the direction is clear: the person licensing your cue wants to know what they're buying, and roughly half of them won't take AI music at all.

That means "is any of this AI?" is becoming a normal question in a pitch, a licence or a library's onboarding form. Being able to answer it quickly, with something to back it up, is now part of being easy to license.

What the EU AI Act actually requires (and of whom)

The EU AI Act's transparency rules, in Article 50, apply from 2 August 2026, according to the European Commission's FAQ on Article 50. Providers of generative AI systems already on the market before that date have until 2 December 2026 to meet the marking obligation. Music Business Worldwide's explainer walks through what this means for the music business.

The obligations fall on two groups:

Providers, the companies that build AI systems and put them on the EU market. Systems that generate synthetic audio, images, video or text must mark their output in a machine-readable way so it can be detected as AI-generated. There's an exception when the AI only performs "an assistive function for standard editing" or doesn't substantially change the input.
Deployers, people or organisations using an AI system professionally. They must disclose deepfakes: AI-generated or manipulated image, audio or video that resembles existing people, objects, places or events and would falsely appear authentic. For "evidently artistic, creative, satirical, fictional" work, the disclosure can be light, made in a way that doesn't spoil the work. Purely personal, non-professional use falls outside these rules.

What it does not do:

It puts no obligation on a composer who doesn't use AI. The Commission's guidance has no transparency duties for people who neither provide nor deploy an AI system.
It doesn't require you to certify that your music is human-made, or to label music that isn't AI-generated.
It doesn't turn "I used a plug-in" into a labelling duty. Ordinary tools aren't the target, and assistive editing has its own exception.

If you do use a generative tool to make audio in a professional context, the Act's deepfake rules matter mainly when the output imitates a real person (a cloned voice, for example). Check the specifics for your case.

The Act also has separate rules for general-purpose AI model makers on training and copyright (a copyright policy, respecting rights reservations, and a public summary of training content). Those are obligations for AI companies, not for you.

Can AI music be copyrighted?

In the US, the Copyright Office's Part 2 report on copyrightability (29 January 2025) is the clearest statement. As summarised by the Office:

AI output is protected only where a human author has determined sufficient expressive elements.
Using AI to assist creation, or including AI-generated material in a larger human-made work, doesn't bar copyright in the human parts.
Creative arrangement or modification of AI output by a person can be protected.
Prompts alone are not enough. In the Office's words, prompts "do not alone provide sufficient human control".

When registering, the Office's registration guidance asks applicants to exclude AI-generated content that is "more than de minimis" and describe it briefly, and not to list an AI tool as an author. The human-only rule held in court: the Supreme Court declined to hear Thaler v. Perlmutter on 2 March 2026.

Other countries have their own copyright laws and the answers differ. The common thread is that human creative choices are what get protected, which is exactly what your working files show.

ASCAP, BMI and SOCAN: AI registration rules

On 28 October 2025, ASCAP, BMI and SOCAN aligned their policies (SOCAN's announcement):

They accept "partially AI-generated" works, defined as works that combine AI-generated musical content with human authorship.
They don't accept works created entirely with AI tools.

The joint announcement doesn't spell out a disclosure form or a royalty split for the AI parts, so check your society's current registration pages. If you're with PRS, SIAE, GEMA, SACEM or another society, look up its own policy; don't assume the North American rules apply.

What to keep to show human authorship

Nobody can prove a negative perfectly. What you can do is keep the trail a normal working composer leaves anyway, and stop throwing it away. Think of it as the creative equivalent of a receipt.

The working files

Project and session files (Logic, Cubase, Pro Tools, Ableton, Reaper, Sibelius, Dorico…), saved under a new name at meaningful stages rather than overwritten.
MIDI, especially performances you played in, with their timing and velocity quirks.
Dated bounces and versions: early sketches, the first mix, revisions after notes, the final, alts and cutdowns. File dates and version names show the music developing over days or weeks.
Stems and multitracks: separated parts show how the track was built.
Recording sessions: dates, studio or room, players, takes. Raw takes are hard to fake and easy to keep.
Scores, lead sheets, voice memos and notebook sketches. A phone memo humming the theme is good evidence of where an idea started.

The paperwork

Collaborator credits and split sheets, with who wrote what.
Sample and loop licences, with the pack name, licence terms and which tracks use them. A sample isn't AI, but "contains samples" is the next question after "is it AI?".
Session musician agreements or invoices showing who played.
The brief and client notes for commissioned work, plus your replies. They show human decisions responding to human feedback.

A short note per track

A few lines written while you work go a long way: when you started, what the cue was for, which instruments are live, which are virtual, and whether any AI tool touched it (and if so, what it did). Writing it at the time is far more convincing than reconstructing it a year later.

The free Human-made music: what to keep (PDF and spreadsheet) turns all of this into a checklist and a per-track log you can fill in. Download the Human-made music: what to keep.

How to state AI use honestly

"AI" covers very different things, and a vague answer makes buyers nervous. Separate three cases:

No generative AI. Say so plainly. Tools like EQ, noise reduction, mastering assistants or sample libraries are normal production, but if you're unsure whether something counts, name it rather than hide it.
AI-assisted. A person wrote and produced the music; an AI tool did a defined job (stem separation, a restoration pass, an idea you then rewrote). Say what the tool did and what you did.
AI-generated elements. Part of the audio or composition came from a generative tool. Say which part. This affects copyright and registration, and some buyers will pass.

Where to state it:

In your metadata or track notes, so the answer travels with the file.
In pitch emails and library submissions, one line is enough: "Written, performed and produced by [name]. No generative AI was used."
In licence paperwork, if the buyer asks for a warranty. Read what you're warranting; ask for wording that matches what you actually did.

Don't overclaim. "100 % human" is a promise; "no generative AI was used to compose or produce this track" is a description you can stand behind.

Answering "is any of this AI?"

When a supervisor asks, they usually want three things: a straight answer, confidence it's true, and reassurance it won't cause trouble later. A good reply covers all three in a few lines:

No generative AI was used on this track. I wrote and produced it in [month, year]; the strings are live (recorded [date] at [studio]), the rest is played in on virtual instruments. I can share the session stems or a work-in-progress bounce if that helps.

If you did use AI:

The composition and production are mine. I used [tool] for [specific job, e.g. separating the drums from an old demo]; no part of the final audio was generated. Happy to walk you through it.

Offering to show the stems or an early version is often enough on its own. Very few people will take you up on it, but being ready to changes how the answer lands.

Your working history in one place, with Aerie

In Aerie, every version of a track (sketches, mixes, alts, cutdowns, stems) lives under that track, and replacing a mix keeps the old one in its history instead of overwriting it. Each track's Sync Passport has a notes field, a "contains samples" field and room to attach paperwork such as split sheets and sample licences, so the answer to "is any of this AI?" sits next to the music. DADAAUDIO never generates music with AI and never trains AI on your music; its AI only listens, to tag tracks and make them searchable.

Frequently asked questions

Do I have to label my music as human-made under the EU AI Act?

No. Article 50 puts obligations on providers and professional deployers of AI systems, mainly around marking AI output and disclosing deepfakes. A composer who doesn't use AI has no labelling duty under it.

Can AI music be copyrighted?

In the US, only the human-authored parts. The Copyright Office says prompts alone aren't enough, but human selection, arrangement or modification of AI output, or AI used as an assistant, can still leave a protectable work. Other countries' laws differ.

Does ASCAP accept AI music registration?

Since October 2025, ASCAP, BMI and SOCAN accept partially AI-generated works that combine AI content with human authorship, but not works made entirely with AI. Check your own society's rules if you're elsewhere.

What's the best proof that my music is human-made?

Dated working files: project sessions, MIDI, early bounces, stems, recording dates and notes made while working, plus credits and sample licences. Together they show human decisions over time.

Should I mention AI in my pitch if I didn't use it?

If the buyer or library asks, or you're pitching to one that cares, yes: one plain line such as "No generative AI was used" saves a round of emails.

Sources

More guides