AI Music: The New Copyright Clusterf*ck?

It’s here: music is having its ChatGPT moment. Or perhaps more accurately, its Midjourney moment. The artificial intelligence train is barreling through the music industry at full speed, a reminder of the tumultuous times of cassette tapes, the fight against illegal downloading and the rise of streaming—now with a metallic AI twist.

As we’ve seen in past times of upheaval, artists, labels and other copyright stakeholders are proactively staking their legal claims in this new territory by filing lawsuits left and right. However, this forward-thinking move could have far-reaching consequences for the music‚ and, particularly, music within the advertising industry.

What’s it all about?

Recently released AI music generators like Udio and Suno can produce remarkably human-sounding compositions in various styles. They can do so because they’re undeniably trained on datasets of copyrighted material from artists, publishers and labels.

Technically, this doesn’t constitute copyright infringement in the traditional sense—there’s no direct copying of melodies, harmonies, specific rhythms or lyrics. The infringement lies in the creation of “derivative work” with the aim of commercial exploitation. In other words, AI music platforms that generate revenue are subject to compensation claims by labels, publishers and artists who argue that creating a derivative work for commercial purposes without compensation is not permissible.

A reality check

Creating a derivative work with the intent to exploit commercially … where have we heard this before? Bingo!

This describes the business model employed by about 50 music-for-media companies in the Netherlands, hundreds in Europe, and thousands worldwide, primarily in the advertising industry, as well as in film and television.

Next to their business model, all music houses, including my old company, have a modus operandi that is identical in processes to Udio and Suno. Reference tracks serve as training data, and the creative brief acts as the prompt indicating how the interpretations should be delivered and what the music should achieve. The composer, or the company, at times with the support of a musicologist, make sure they avoid traditional copyright infringement.

If an AI is not allowed to create derivative works to replace an originals for financial gain, why should a company with real people be allowed to do so? And who’s to say the same labels, publishers and acts won’t also retroactively pursue these companies, composers and artists for compensation?

One step further

If we take this notion one step further, what does it mean for artists and bands in general? Hell, what does this mean for artistic creation in the broadest sense possible?

I was in a band called Das Oath. Arguably, a group considered “authentic” and “real” that created “original” works. The reality is that for our sophomore album, we ripped off the entire In Utero record by Nirvana (training data) and just played it mega fast and aggressive (prompt). Do we retroactively owe Frances Bean Cobain money?

Hopefully, new laws will soon bring clarity to this legal tangle. For now, it remains a complex playing field with many interests at stake. It’s a realm where the definitions of “original” and “derivative” must be oh so clearly stated.

I’m pretty sure Frances Bean will leave me be. However, if I were still running a music agency whose production process consists of briefings stating: “Take these five references, copy them, but make them different enough so we don’t break any laws,” I’d start to worry.

There’s no telling what copyright infringement will look like tomorrow.