This version is no longer in force.

It took effect on 2026-09-26 and was superseded on 2026-10-11. It is kept here unchanged so that anyone who needs the version in force on a past date can cite it.

Version of 2026-09-26

AI transparency notice

Euphona processes audio with AI, and the law requires us to say so in two ways: a machine-readable mark on what comes out, and a plain notice where you use it. This page explains both, and is honest about the part that is still with our lawyers — which of our capabilities count as 'substantially altering' your recording. Superseded by the 2026-10-11 corrections, which changed no practice and narrowed no right: it restated Article 50 of the EU AI Act accurately — which paragraph asks for what, that the editing exemption lifts only the marking duty, and when a disclosure is due — and called the vocal remover by what it is.

What this covers

Article 50 of the EU AI Act has applied since 2 August 2026. It requires that audio which has been artificially generated or manipulated is marked in a machine-readable way, and that people are told when they are interacting with an AI system. Euphona is an audio processing platform, so both parts apply to us rather than to only one of our surfaces.

This notice is the durable reference for how we meet them. It is not, by itself, how we meet them — a transparency duty discharged only by a policy page is a duty not discharged, and we treat it that way internally.

Marking what comes out

Audio that Euphona has generated or substantially altered carries a machine-readable mark identifying it as such. Three things about that mark matter more than its format:

It is applied in the render and export path, so it travels with the file rather than with the page you downloaded it from.
It has to survive the chain. A mark applied before a transcode that the preview or export step strips has achieved nothing, so surviving the pipeline is part of the test, not an afterthought.
It identifies the audio as machine-processed. It is not a watermark on your music, it does not encode who you are, and it is not a restriction on what you may do with your file.

The Act allows for what is technically feasible. Where a format or a delivery path cannot carry the mark, we record why rather than assuming the allowance covers us.

Telling you where you are

The second duty is disclosure at first exposure: you should know you are using an AI system at the point you start using it, not in a document you never open. So the notice appears in the Studio and in the free tools at first interaction, in the interface itself.

This page exists so there is somewhere permanent to read the detail. It is deliberately not the mechanism.

Which capabilities are in scope

Article 50 exempts systems that perform an assistive function for standard editing, or that do not substantially alter the input data. Our capabilities do not all sit on the same side of that line, and we are not going to pretend otherwise:

Mastering, restoration and mixing process a recording you made. They plausibly fall inside the exemption, because the material is yours and the work is editing.
Stem separation and vocal tuning are harder. Separating a mix into parts, or re-synthesising a vocal, can alter the material substantially enough to fall outside it.
Anything that generates audio rather than processing yours is plainly in scope.

Which side each capability lands on is a legal determination, and it is our lawyers’ to make rather than ours to assert. That analysis is in progress and, as of the date at the top of this notice, has not been concluded for any capability. Until it is, every capability is treated under the rule below, so what we publish about a capability and what we have determined about it cannot disagree.

Where a capability’s status is genuinely uncertain, we mark and disclose. Over-marking costs a line of interface; under-marking costs the one claim this company is built on.

What is true today

All six capabilities — mastering, stem separation, audio restoration, vocal tuning, mixing and spatial audio — are available in the Studio, and the free vocal remover and the product-page demos on this site run the same engine. Every file any of them writes carries the mark described above: it is written into the deliverable and into the preview you hear when they are made, and the pipeline checks that it is still there after every conversion before anything is handed over. A job whose output cannot be verified as marked fails rather than delivering an unmarked file. The disclosure appears at the result, in the Studio and on the tool pages, before any download.

Which side of Article 50 each capability falls on is a determination for counsel, and none has yet been taken. Every capability is therefore marked and disclosed under the rule for uncertain cases rather than left to wait for the answer, and each will be treated that way until its own determination says otherwise.

This paragraph has a shelf life, and that is the point. It names what has shipped and what has been determined, and it is updated as part of releasing a capability, alongside the quality bar it has to clear and the counsel determination on which side of Article 50 it falls. A page that is accurate on the day a capability ships and wrong a week later is the failure mode this notice exists to avoid, and it is the easiest one to fall into because nobody has to do anything for it to happen.

Questions and corrections

If something here is wrong, unclear, or does not match what you observe in the product, tell us at legal@euphonaai.com. A transparency notice that does not match the product is worse than none, so a correction is the most useful thing you can send us.