Music, Artificial Intelligence and Copyright: A debate that can no longer waits

Can a song generated by an algorithm have an author? Who protects artists when an AI trains on their music without asking permission? I contributed to a feature in 20minutos that tackles these questions head-on.

Artificial intelligence has been composing music for a while now. Not as a lab experiment, but on Spotify, in real playlists, competing alongside human artists. Bands like The Velvet Sundown and Breaking Rust — entirely generated by AI — already have millions of streams. And behind all of this lies a question that the law still does not know quite how to answer.

I had the opportunity to contribute to a feature published in 20minutos, written by Sofía Puvill Balcells, which explores in depth the current state of AI-generated music: what tools exist, what artists think, and, above all, where the legal framework currently stands.

The authorship problem

By legal definition, a work is the result of an original human intellectual creation. When an AI model generates a song from a prompt, that condition starts to crumble. Human involvement may be minimal or entirely indirect, and that opens a crack in the entire intellectual property system as we know it.

What truly matters from a legal standpoint is not how the result sounds, but how it was produced. The law does not distinguish by outcome, but by the origin of the work: the degree of genuine human involvement in the creative process. When an artist freely decides the structure, melody, arrangements and final form of a piece, authorship exists. When the algorithm does, it does not.

Training Data: A dangerous legal gap

One of the most sensitive points raised in the feature concerns training data. AI music models are fed large catalogues of songs. If those songs are protected by copyright and their authors have not given consent, we may be looking at infringement. The problem is that proving it is extraordinarily difficult.

Artists have no access to the model or its training datasets. The opacity of these platforms makes defending one’s own rights a near-impossible task without judicial intervention. European law does offer mechanisms to compel the disclosure of information when there is well-founded suspicion, but the process is long and costly.

Towards a new model: Licensing, transparency and a dual system

In the feature I argue that if algorithms train on other people’s works, they should obtain a licence to do so and pay a royalty to the artists. This position rests on a basic principle: whoever profits economically from the creative work of others must compensate them.

I also propose moving towards a dual copyright system: one for works with genuine human authorship, preserving classical copyright, and a separate framework for AI-generated content. Both should be accompanied by clear transparency obligations, so that listeners can know at all times who — or what — created the music they are hearing.

The full feature also includes the voices of artists such as Maria Arnal and bands like VITTARA, along with reflections on streaming platforms and the economic impact AI may have on copyright revenues in the coming years. I invite you to read it here:

AI-Generated Music, Between Opportunity and Legal Grey Areas — 20minutos

If you have questions about the protection of AI-assisted works, copyright in the digital environment, or any intellectual property matter, feel free to get in touch.

— David Moreno, Intellectual Property Lawyer and Founder of CopyrightStudio