MBW Views is a series of op-eds from eminent music industry people… with something to say. The following MBW op/ed comes from Walter De Brouwer, a generative linguist by training and a co-founder, alongside Michael Ovitz, of SoundPatrol. He also teaches at Stanford.
SoundPatrol, described as a research lab for large music models, struck a “first-of-a-kind collaboration” with Universal Music Group and Sony Music in September 2025 to protect artists from copyright infringement activity coming from AI music generators.
In this op-ed, De Brouwer argues that a catalog is no longer just a collection of songs; it’s becoming raw material that machines draw on. As that happens, the market splits in two: cheap, infinite AI music on one side, and rare, verified human work on the other.
His proposed fix is the “transformation royalty” – a payment that reaches the artist the moment a machine generates something from their voice, style, or songbook…
Leonard Cohen took five years to write “Hallelujah.” He wrote eighty verses and threw away most of them. Last month, a software model trained on a few thousand songs like his generated a Cohen-adjacent ballad in nine seconds.
Cohen’s song sits inside a $4 billion catalogue Sony Music Publishing just bought from Blackstone, alongside “Don’t Stop Believin’,” “Umbrella,” “Bad Romance,” “Single Ladies,” and forty-five thousand others. The trade press read it as a vote of confidence in nostalgia. It wasn’t. It was something stranger, and more honest. It was a bet that in five years, the most valuable thing a music company can own will not be Cohen’s song. It will be the right to license what a machine can do with Cohen’s voice.
The catalogue is becoming a function, not a file. And almost no one outside the executive suites is talking about what that means.
Meanwhile, Grimes did something different. In April 2023 she opened her own voice to the generators and said: split the royalties fifty-fifty, generate whatever you want. She was treated, at the time, as eccentric. In retrospect, she was the only person in the room who had read the room.
What the majors are actually doing
The major labels have a way of dealing with new technology. They treat it like a teenager. First the walled garden, then the curfew, then the lawsuits (the “grounded stage”), then the licensing deals that look like college tuition. They did this with radio, with cassettes, with MTV, with Napster, with YouTube, and with Spotify. Every single time, the new technology was going to destroy music. Every single time, the labels’ real strategy was to observe its behavior, set the rules, and eventually send it off into the world as a paying customer.
The strategy was decided years ago. The lawsuits are how the boundaries get drawn. The licenses are how the relationship gets formalized. And the kid eventually leaves home, gets a job, and starts paying rent in the form of royalties on every model output that derives from a licensed catalogue.
The industry has spent two years arguing past this entire process. One camp says generative music is theft and must be stopped. The other camp, well-funded, mostly West Coast, fluent in the language of innovation, says the technology is going to ship regardless – and the lawyers should get out of the way. Both camps are arguing about the wrong thing. Neither is wrong about their values, but both are wrong about the clock.
The clock is the actual antagonist. The legal questions take years. The technology ships in weeks. By the time any of the current cases produce a binding precedent, the supply of generative music will have grown by orders of magnitude, and the precedent will apply to a market that no longer looks anything like the one the case was filed about.
The head fractures
For a hundred years the music business ran on scarcity. A small head of hits drove most of the revenue, and the engine was simple: protect the hits, ignore the rest. Streaming did not break that. It made the head bigger. Drake alone accounts for more of Spotify than the bottom fifty thousand artists combined.
Generative music breaks it for real, but not in the way the doomsayers think. It is not that the tail swallows the head. It is that the head splits in two: one half becomes a cheap, infinite, adaptive utility, and the other half becomes a luxury good built on verifiable human origin. This is what the Blackstone-to-Sony deal is really about: The buyers are not betting on nostalgia. They are buying two assets in one envelope: the human-attested core that will become the luxury good, and the rights footprint that will license the machines that generate the utility. They are options on both sides of a fracturing market.
Generative listening
There is a name for what is coming on the consumer side, and almost no one is using it yet. Call it generative listening.
The recording, as a unit of consumption, is about to disappear. What replaces it is a continuous, personalized rendering: your favorite song, sung by Karen Carpenter, with Prince on guitar, at the tempo of your run, ending exactly when you reach the door. Whitney Houston covering a track she never lived to hear. Jimi Hendrix soloing over a song written last week. The dead artists you love, doing the work they never got to do. The living artists you love, doing more of it, in versions that fit your day.
For most of music history, listening was passive: you received what had been recorded. Generative listening reverses the direction. You listen first; the music is produced in response. The performance becomes the function call, and the catalogue becomes the licensed library of voices and styles the function draws from. Karen Carpenter recorded a finite number of songs in her lifetime. In generative listening, her voice becomes a renewable resource, and her estate, if the rails get built, becomes a perpetual royalty engine instead of a slowly depreciating asset.
A Deezer/Ipsos study found 97 percent of listeners cannot tell AI music from human music in a blind test. That is not a tragedy, it’s a market structure. When the difference is undetectable, the difference becomes the product. The unmistakable signal of human origin – the live take with breath and imperfection, the limited drop, the song you know was written by a person who spent five years on it — becomes the scarce thing, and scarce things command premiums. Vinyl is selling at numbers not seen since 1988. The same logic is about to do extraordinary things for human-attested music.
This is also, finally, the thing that resolves the question of what the human-attested core is *for*. It is the seed. The live take, the original session, the verifiable human performance – these are the source material that the listener’s generative session draws from. The luxury good and the utility are not two separate businesses. They are two ends of the same transaction. The luxury good is the asset. The utility is what the asset becomes when you press play.
The transformation royalty
Here is what the next five years actually look like.
Every time a model generates something derived from your voice, your style, your face, your songbook — a micro-payment routes to you in real time. Not a settlement years later. Not a class-action coupon. A royalty on every transformation, settled the way credit-card transactions settle, riding on payment rails that already exist.
Call it a transformation royalty. The labels and the platforms are already moving toward it whether they say so or not. Universal licensed Udio. Warner did the same with Udio and Suno. KLAY signed with all three majors. The lawsuits were the boundary-setting. The licenses are the business. The teenager is leaving for college.
The artist gets paid. The listener gets a personalized version of music they actually want. The labels finally get a model that scales with usage rather than fighting it. The piracy era did not end because the industry sued every fan. It ended when authorized listening became cheaper than the friction of theft. The generative era ends the same way.
What the artists who get it first will do
Grimes was early. So was Holly Herndon, who has been building consensual voice models since 2021. Both understood something most artists are still catching up to: The question is no longer whether the machines will generate music in your style. They will. The only question is whether you have a contract with them.
The artists who win the next decade will be the ones who set the terms before the terms are set for them. Open your style, but on your contract. Sell the transformations but ring-fence the live performances and the limited drops. Treat your voice the way Supreme treats its drops — verifiable, scarce, expensive when it counts. Let the machines do the workout playlists. Keep the church gig for yourself.
Leonard Cohen will never sign a transformation royalty agreement. He died in 2016, before the question existed. But the song he spent five years writing is now an asset in a $4 billion portfolio, and somewhere a model is generating a Cohen-adjacent ballad in nine seconds for a listener who will never know the difference.
The question for the living artists is whether they want to be paid every time it happens. Not with lawsuits. With a price.Music Business Worldwide
