A fun consideration. From the distant perch of a forensic musicologist at least. And for now, at least.
I just finished reading The National Law Review’s “Universal Music Group May Have Admitted “Dilution” Theory in Copyright Infringement Case Versus Suno Is Baseless.” Fun at first because I guess it’s a bit of a gotcha. On March 6, UMG’s Chief Digital Officer Michael Nash gave investors a reality check on the Q4 earnings call. AI-generated content “barely registers” in the marketplace—less than half a percent of organic consumption. The case for AI royalty dilution, he said, has been “massively over-extrapolated from a few anecdotes.”
It’s one thing to be speaking to your investors and quite another speaking to a court, but let’s not oversimplify for the sake of gotchas. The problem is that UMG is simultaneously telling a federal judge the exact opposite in their suit against Suno. In that courtroom, the theory is that AI-generated music will “overrun” their catalogue, “compete for plays against real, copyrighted sound recordings,” and “substantially dilute the royalty pools” paid out to artists.
Do we have to pick one, or can we keep the now and the future in our heads at the same time?
Or is this partly gamesmanship, and that’s not at all the question for me to be asking. Fun.
You cannot tell a judge that AI is an existential threat to your revenue while telling Wall Street your revenue is fine and not expect to hear about it, perhaps at deposition. At a March 12 hearing, Suno’s counsel read Nash’s comments into the record. UMG tried to block discovery into the remarks and the data behind them. Judge Levenson said no—a bad day for UMG. (I think I’m finished with not using em’s by the way. We’ve gotta get over it.)
But Nash’s comments didn’t kill dilution theory if it was already dead.
Dilution theory posits that if AI users can produce enormous volumes of new music, the value of existing copyrighted works gets diluted. It is, as Professor Edward Lee of Santa Clara Law has called it, “sweeping.” If I have it at all straight, it proposes to treat every output of every AI generator as dilution of every copyrighted work used in its training. That’s a big wish, given copyright law’s fundamental limitation to substantially similar copies.
I keep trying to play lawyer and I find myself thinking copyright law both wasn’t at all prepared specifically for AI (how could it be?) and yet accounted for it just fine, and was in advance unconcerned. But hey, I could be wrong. I just don’t seem to be.
Judge Alsup rejected dilution outright in Bartz v. Anthropic, writing that even an “explosion of competing works” is “not the kind of competitive or creative displacement that concerns the Copyright Act.” The Act exists to advance original works of authorship, not to shield authors from competition. The U.S. Copyright Office acknowledged that dilution enters “uncharted territory” with no basis in existing case law, though the Office endorsed exploring the theory—a position that itself has drawn criticism from scholars like Prof. Lee, who argued the Office overstepped on a theory no court has adopted. Judge Chhabria entertained the idea in Kadrey v. Meta, but the comments are dicta which I understand to mean a conversational aside, not something to be upheld.
So we have a theory with no statutory basis, no case law precedent sort of thing, one judge against it on the merits, another judge’s non-binding dicta just a little more open to it, and the plaintiff’s own high up said the harm is nothing to sweat. I suspect he meant just not yet. He was talking to investors. I’m really pretty sympathetic here.
This is more my take. Copyright law has never protected anyone from competition. That’s not what it does. Every new song that ever comes around dilutes the value of the ones that came before it in some microscopic way. It’s supply. The entire history of music is one long chain of new sounds displacing old ones. Motown didn’t sue Philly. Funk didn’t sue Disco. And R&B didn’t sue Hip-hop. Displacing revenue from the foregoing is progress. Copyright protects specific expressions. Nothing to do with protecting market share. When UMG argues that Suno’s output “competes for plays” against its catalog, they’re describing competition. That’s shaky.
Professor Brian Frye at the University of Kentucky cuts through the noise on this, describing the labels as “landlords” who’d like to go on collecting rents. The labels aren’t arguing their specific songs are being copied. They’re still in the macro where everyone’s songs sorta were and sorta weren’t. That’s gonna pass. But right now that’s so core to this, everything else is reliant upon, stuck on, feelings about training. But they’re arguing the market for music should stay the size and shape that benefits them. Copyright as far as I can tell doesn’t get involved with that.
I was told there’d be no research.
Article I, Section 8, Clause 8 states its own purpose: existing “to promote the Progress of Science and useful Arts.” To promote progress. If 60,000 new tracks a day is what progress looks like in 2026, the law’s job appears to be to stay out of its way.
I shoehorned that in there.
Sheet music publishers tried to stop the player piano. The record labels tried to stop Napster. Every one was met with the same existential panic from the same type of incumbent. Every time, the law accommodated the technology because the public wanted what it offered. The compulsory license wasn’t a victory for sheet music publishers. It was a structured surrender. The DMCA wasn’t a victory for the labels—they killed Napster and it didn’t matter. Congress just built guardrails around a reality they couldn’t reverse.
Enter YouTube and Spotify.
None of this means generative AI presents no legitimate copyright concerns. It does. As it always was, if a model produces output that is substantially similar to a specific copyrighted work reproducing protectable expression from an identifiable source, call me. Finding the lines between common musical vocabulary and specific authored expression is the work of forensic musicology. That work becomes more important, not less, as the volume of new music increases. The opportunities for both infringement and false accusation obviously multiply.
That’s copyright. But dilution theory isn’t about specific works being copied. It’s about the market getting bigger than the incumbents are comfortable with. I don’t blame them one bit, except that I don’t agree. I’m still going to want to hear the next Bruno Mars record. Maybe I’m too sanguine. I’m still thinking it through. But copyright law doesn’t owe any industry a fixed supply.
I’m thoroughly amused by the debate, but eventually, I expect all this litigation will converge upon that reality.
We’ll return to where we were, focused on the questions that actually matter to copyright.
