Written by Editors Note: This is part 1 in a three part series on AI in the music business.
Having spent fifty years in this business, I’ve been witness to most of the seismic changes this industry has seen. That in itself will be the subject of a separate rant later this year, but a recent Wall Street Journal article, reporting that the RIAA and IFPI are pushing streaming platforms to label AI-generated tracks, spurred me on to get something off my chest. Over the past year I’ve been hearing the same alarm about AI in almost every conversation I have with musicians, and I understand it. It is personal, especially generative AI, and their instincts are not wrong to be wary.
Where I think many of them go wrong is in where they point that fear, and it’s worth separating two arguments that keep getting collapsed into one. Labels have sued platforms like Suno and Udio for copyright, arguing their models were trained on unlicensed recordings at industrial scale, and a number of independent artists and third parties have taken up the same fight in class actions of their own, mostly because they don’t trust the major-label settlements to protect anyone but the majors.
As of last week, part of this has an answer. A Munich court ruled that Suno infringed copyright by training its models on six specific songs, among them Daddy Cool, Rasputin and Forever Young, represented by the German collecting society GEMA, without ever obtaining a licence. Suno was ordered to disclose the revenue tied to that infringement and will owe damages, still to be calculated, and can appeal. It is the first major European court ruling on AI music training, and the reason it’s an important ruling is that a court just confirmed, song by song, that a commercial product had reproduced identifiable elements of specific copyrighted recordings it never paid to use.
Now this is a legitimate concern, and artists are right to care about it, in principle!
It’s worth knowing, and it’s the kind of detail I find darkly ironic, that copyright was never built to protect artists in the first place. In its earliest form, copyright emerged as a printers’ privilege, not an author’s right: after the printing press spread through Europe in the 15th and 16th centuries, England’s Stationers’ Company was chartered in 1557 and given a government-backed monopoly over printing and bookselling, largely as a tool for the Crown to enforce censorship, with the “stationer’s copyright” simply preventing rival guild members from reprinting each other’s editions.
It took a hundred and fifty years, until the Statute of Anne in 1710, before Parliament bothered to give the author any legal standing at all, and even then the act protected “authors or proprietors” in the same breath, which mattered a great deal in practice, since most authors had already sold those rights straight to a bookseller, meaning the law dressed up as protecting the writer mostly ended up protecting whoever already owned the printing press. Which makes it worth asking exactly whose interests are being protected when Universal and Sony, two of the largest corporations in recorded music history, drag another corporation into court over training data. I’m sure the artists on their rosters are touched by the concern. Copyright has always been a tool for whoever controls the printing press to fight over who else gets to use it, and for three hundred years that’s rarely been the person who actually wrote the song, whatever the paperwork says.
But there is a second, different fear riding alongside all of this, and this is the one I think is being sold to musicians as something it isn’t. It goes something like this: AI is coming to replace you, to impersonate you, to make you obsolete, and training on your music is basically theft of who you are. Here is where I want to push back, because these are two entirely different claims. In my younger days I was trained, like every music student on earth, on Bach and Beethoven, and none of it particularly moved me. What actually got under my skin was Led Zeppelin, Jimmy Page and John Bonham specifically, and neither of them asked permission before doing that to me. Nobody asked Bach’s estate for permission either, and nobody owed him a royalty for it, because absorbing influence over a lifetime, whether it’s the theory you were made to learn or the record that actually changed how you heard music, and building a commercial product by reproducing someone’s specific, identifiable work are not the same act.
The GEMA ruling wasn’t about Suno being generally inspired by six songs the way I was inspired by Jimmy Page and John Bonham, it was about a court finding that prompted outputs reproduced recognisable melodies and harmonies from those specific recordings, without a licence, inside a product Suno charges for, and there in lies the point!
That is the legitimate breach of copywrite, but it has nothing to do with identity theft. No platform serious about staying in business is going to hand a user a convincing deepfake of a named, working artist, that’s is a lawsuit and a banned platform waiting to happen, and every major streaming service already prohibits it outright. The identity-replacement fear is largely noise. The unlicensed-reproduction fear is real, and I think the labels are perfectly happy to let the noisier, scarier version dominate the conversation, because it keeps the villain fixed on the AI company and away from the one place the labels themselves are quietly far more exposed: the royalty pool their entire business now depends on.
Now don’t get me wrong, I’m not advocating for “AI slop” here, I’ll get to that later in this piece. I’m just trying to highlight a point my bullshit meter has been flagging for some time now. Generative AI may be the first real threat, in twenty years, to the people who have actually been profiting from streaming, and that is not musicians.
Streaming did one thing very well over the past decade. It opened up distribution and discovery that a musician of my generation could barely have imagined, a recording made in a small studio in Rotterdam reaching a curious listener in Osaka within a day of release, and I do not want to pretend otherwise. I will come back to why that matters for jazz specifically later in this series, believe it or not, it’s more significant than most people in this industry realise.
But streaming did something else at the same time, and it is worth being precise about what actually changed. Now while a musician’s masters and publishing are governed by the same contracts and the same rights structures that always existed, what changed was the way music was consumed, and this shifted the balance in the earnings model decidedly more in the favour of the labels.
What exactly changed was the listener’s relationship to the music. A generation ago, a fan who wanted an album bought it, a single transaction, a copy they then owned outright. Streaming replaced that purchase with a subscription, a monthly rent paid for access to everything at once rather than ownership of anything in particular, in other words all you can eat for ten bucks a month. That single shift, from buying a copy to renting access, rebuilt the entire economics of recorded music around a shared pool of subscription revenue divided pro rata across every stream on the platform. And in a pooled, pro-rata system, scale wins by something close to mathematical necessity. The companies holding the largest catalogues, which is to say the major labels, capture an outsized share of that pool almost automatically, while the independent working musician, however loyal their following, is left receiving a fraction of a cent among an enormous shared total. The industry simply built a payment system that rewards catalogue volume, and the companies that already owned the largest catalogues were always going to be the ones who benefited most.
This is the part of the current AI panic that I think most working musicians are missing, and it isn’t the part the German court just ruled on. Let the labels have their courtroom. Watching two of the biggest companies in recorded music fight over training data is corporate theatre, and it’s not particularly your theatre.
Here’s the part that might actually surprise you: the royalty pool argument isn’t yours to worry about either, not the way it usually gets framed. The overwhelming majority of working musicians are already earning next to nothing from streaming, a fraction of a cent divided across a pool that was rigged against you from the start, so a threat to that pool’s integrity doesn’t cost you much you didn’t already not have. The party with something real to lose is the landlord, the major labels sitting on the largest catalogues, who depend on that pool staying large, stable and trusted, because that’s how they keep capturing their outsized share of it. Every fraudulent AI stream draws down the exact same pool they’ve built their entire business on protecting. You were never winning the volume game. They were. Which means, for the first time in twenty years, the party most exposed by any of this isn’t the person on the bandstand, it’s the one who’s been quietly landlording over your royalty statement the whole time. And it hands the independent artist something the pool was never built to reward: a direct, provable claim to being real, at exactly the moment listeners are actively hunting for that. There’s a decent joke buried in all this, if you’re in the mood for one. When the labels took the recorded-income side of the business for themselves, musicians didn’t sit around waiting for a fair deal, they went back to the stage, because that was the one income stream no landlord could put a lease on. AI slop has the audience running for the exits, and where they’re running to is exactly the room musicians never actually left. The labels built a wall around the recording. They forgot the gig was never inside it.
Let me put some numbers behind why the anxiety exists in the first place, because the volume is genuinely startling. AI-generated tracks now account for as much as 50 percent of all daily uploads to a platform like Deezer on peak days, up from 44 percent just two months earlier, with bulk uploaders pushing nearly 90,000 fully AI-generated songs into circulation every single day, worldwide, without pause. That is not a trickle. That is a flood, and it has been rising for well over a year with no sign of slowing.
But volume of uploads and volume of actual listening are two entirely different measurements, and the industry has been remarkably slow to separate them. On Deezer, the one platform that has published real numbers rather than talking in generalities, AI-generated tracks capture just one to three percent of total streams despite making up nearly half of daily uploads. And of that small slice, Deezer says up to 85 percent of the plays are fraudulent, run through botnets built to siphon money out of the royalty pool rather than to reach an actual human ear, against a fraud rate across Deezer’s entire catalogue of roughly 8 percent. This is not a music discovery problem. It is a plumbing problem, and on the evidence of the one platform willing to show its working, the plumbing has been rigged.
So on the surface, the raw numbers look alarming, and I understand why they’ve got musicians rattled. But upload volume and fraud statistics only tell you what AI companies and bad actors are doing. They don’t tell you what listeners actually want, or whether audiences can tell the difference at all, and that turns out to matter more than any of the numbers above, and it turns out to have a lot to do with where I said musicians already ran to. That’s where I want to take this next, in part two of this train of thought, which I’ll publish shortly.
Featured Image by Claude Sonnet 5 and prompted by the author.
Last modified: August 18, 2026
