There are two new metadata fields on a SoundCloud upload as of this month. The first is labelled TRAINED ON. The second is labelled TRAINED OFF. The first is a checkbox. The second is a list.

The TRAINED ON checkbox indicates that the artist has consented to the track being included in training data for music-generation models. The TRAINED OFF list is the artist’s opt-out: a comma-separated set of model names or platforms the artist explicitly does not want this track touching.

You can fill out one. You can fill out the other. You can fill out both. You can fill out neither. The platform’s default is neither, which is itself a meaningful position.

What I want to talk about is what these fields are doing, because it is not what the public conversation has assumed. The public conversation has read the fields as a copyright tool. They are not, exactly, that. They are a self-classification.

Consider the difference. A copyright tool would assert ownership and force compliance: this work is mine, and any model that trained on it without my consent owes me. A self-classification tool says something quieter: this is the position the artist takes on the question, and you can choose to honour it.

The field is, in this sense, more like a content rating than a take-down notice. It is information for the platform and for the listener and, eventually, for the model trainer. It does not, by itself, compel anyone.

Whether the trainers honour the field is the open question of 2026. Some of them have committed publicly to respecting opt-outs. Others have not. The legal architecture under which any of this is enforceable in any jurisdiction is, kindly, in flux.

But I am interested in what the act of filling in the field does, regardless of whether it is enforceable.

It is a position-taking. It is the artist saying, in public, what they think about training. The field is small, and the act is small, and the cumulative weight of millions of artists doing it is, slowly, an industry-wide answer to a question the industry has been refusing to answer in any other format.

The pattern that has emerged in the first month is this: established artists with catalogues are largely TRAINED OFF. Newer artists who are still building audiences are split. Some of them are TRAINED ON because they want their work to be inside the conversation, even at the cost of training. Some of them are TRAINED OFF because they have read the room and have concluded that opting in is the same as agreeing to be quoted out of context forever.

Both positions are coherent. Both are, in a small way, surprising. The discourse had assumed the older artists would be more permissive — more confident in their position — and the newer artists would be more protective. The pattern is the inverse. The older artists are protecting the catalogue; the newer artists are gambling on visibility.

There is a third position that has surfaced and is worth noting. A small number of artists have, instead of choosing one box or the other, written a note in the description field that reads something like: trained on with attribution, trained off without. They are, in effect, drafting a licence in plain text on the upload page. The platform did not ask for this. It is happening anyway.

This is what self-classification turns into when the platform does not provide enough granularity. The artists, given a checkbox, write a paragraph. The paragraph is, for now, a marker — a signal to anyone who reads the description that the artist is paying attention. Whether anyone reading the description will honour the paragraph, again, is the question.

The metadata field is doing more work than the platform admits. It is, quietly, organising the field into positions. The next year will tell us whose positions the trainers respected, whose they did not, and whether the difference matters in practice. The field, for the moment, is the closest thing the music industry has to a public register of consent. That is not nothing.