I made seven songs with AI. Who is the author?

This site now has a music player with the AI Radio songs. They are about tokens, hallucinations, always-on agents. There are seven. I did not write them: I explained what I wanted to achieve, the idea to tell, the message, the mood and the style, starting from the site’s guides. Everything else, lyrics, melody, arrangement, instruments, vocals, was made by an AI model in a few minutes per track.

Let me say it up front, because it is the point: these are real songs, pleasant to listen to, and no musician worked on them.

What surprised me

I described the style in one line (“indie rock, gritty guitars, raspy vocals, guitar solo”) and got two complete versions, with verses, chorus, bridge and ending matching the idea I had explained. The model knew where to put the emphasis, when to drop the volume, how to close.

For content creators it is a quiet revolution: a podcast theme, a shop jingle, the soundtrack of a company video no longer cost weeks and hundreds of euros.

What is missing

And yet, listening again, you can hear what isn’t there. No risky choices, no happy accidents, no story behind the voice. They are songs that are average in the statistical sense: the perfect blend of thousands of similar tracks. It is the same limit I see in generated text, which I discuss in AI slop and judgement.

The part that feels mine is the intention: knowing what I wanted the listener to understand and picking, among the versions, the one that said it best. The rest is a well-tailored suit, made to measure for nobody.

So who owns them?

Here the answer is less romantic and more legal.

In Italy, Law 132/2025 clarifies that copyright protects works that result from human intellectual effort, even when made with AI’s help. Since I only gave the direction and the AI generated both lyrics and music, these songs are most likely not protected by copyright.
In the United States the Copyright Office takes the same line: what the machine produces on its own cannot be protected; the human creative contribution is what counts.
Then there are the terms of service: on Suno’s free plan use is non-commercial only; using tracks for a business requires a paid plan.
And there is the upstream question: which songs were these models trained on? Record labels sued music generators and the industry is moving towards licensing deals. The story is not over.

Transparency, always

These songs don’t imitate any real person’s voice, so they aren’t deepfakes and the AI Act’s disclosure duty (Art. 50) doesn’t automatically apply. I disclose it anyway, by choice: listeners have the right to know that the singing voice belongs to no one. That is why the player and the Music page clearly say “AI-generated music”.

My position

Using AI for music is not cheating, if you say so and add something of your own, even just a clear idea of what you want to say. The real risk is different: that ease takes away the desire to learn to play, to write, to get it wrong. The site’s seven songs are a fun experiment and a useful way to explain. They do not replace the evening when a friend picks up a guitar and misses a chord.

In one line

AI can make songs. The songs that matter are still the ones where someone has something to say.

Sources

Informational content, current as of the date shown. Prices, models and rules change often: always check official sources. Written with the support of AI tools and reviewed by the author.