Onpode
Cover art for When AI trains on copyrighted culture, whose creative rights disappear?

When AI trains on copyrighted culture, whose creative rights disappear?

November 20, 2026 · 9 min

Cole Brennan & Malcolm Reeves

Long one or short one this week for you? Long. Honestly couldn't stop reading about this — I kept finding new plaintiffs every time I refreshed. Now, that's a good sign we picked the right week to get into it. Here's where I want to start — not with the lawsuits in aggregate, but with…

A wave of copyright lawsuits has accumulated against AI companies—most prominently Google, Anthropic, and Suno—over the alleged use of copyrighted cultural works to train generative AI systems, without licensing, payment, or attribution.

0:008:34
Get the next episode on Culture →

Follow it free — new episodes land in your feed.

Or make your own — any topic, in minutes

More Onpode episodes on Culture →

About this episode

A wave of copyright lawsuits has accumulated against AI companies—most prominently Google, Anthropic, and Suno—over the alleged use of copyrighted cultural works to train generative AI systems, without licensing, payment, or attribution.

Grounded in 9 sources
Anthropic, music publishers square off on ‘fair use’ defence in AI copyright case | The Straits Times ↗ · straitstimes.com
Hollywood backs NYT in AI copyright fight with OpenAI and Microsoft ↗ · yahoo.com
cnet.com ↗ · cnet.com
theverge.com ↗ · theverge.com
AI Training Data Copyright: Fair Use and Licensing ↗ · astraea.law
Music Industry Challenges AI Training in OpenAI Case | AUDIARTIST ↗ · audiartist.com
Music industry intervenes in Open AI’s big legal battle with newspapers and authors ↗ · completemusicupdate.com
All Shook Up — Elvis Presley's Rights Holders Sue Anthropic Over Lyrics Used to Train Claude ↗ · foxrothschild.com
Irving Azoff's MAC enters OpenAI copyright fight | The Goldmidi ↗ · goldmidi.com
Read transcript

Malcolm Reeves: Long one or short one this week for you?

Cole Brennan: Long. Honestly couldn't stop reading about this — I kept finding new plaintiffs every time I refreshed.

Malcolm Reeves: Now, that's a good sign we picked the right week to get into it. Here's where I want to start — not with the lawsuits in aggregate, but with one number. September 2026, Sony Music Entertainment and Universal Music Group file a follow-on complaint against Suno, and the allegation is 60,202 copied recordings. That's what they're claiming. And the reason that number stops me is — it's not a rhetorical estimate. Someone did an actual audit of what Suno trained on and produced a specific count.

Cole Brennan: So that's — wait, that's the largest single copyright complaint ever filed?

Malcolm Reeves: On record, yes. And it lands inside a much larger pattern — at least 145 copyright infringement suits filed against AI companies by the time we're talking about this. The New York Times against OpenAI and Microsoft, December 27, 2023. Reservoir Media filing two separate suits on October 21 against both Suno and Anthropic, claiming unauthorized use of over 1,050 songs. This isn't a few aggrieved parties. It's a coordinated, industry-wide push.

Cole Brennan: And Reservoir went after both Suno and Anthropic separately — same day. That's deliberate.

Malcolm Reeves: It is. And the thing none of these suits have done yet is settle in any way that changes how these platforms operate. Which is the part I genuinely don't have a clean answer for — why does this keep widening instead of resolving?

Cole Brennan: Yeah, that's the thing. And I think — I mean, some people would say the companies are just running out the clock, but I'm not sure that fully explains it either.

Malcolm Reeves: But running out the clock doesn't explain what Judge Alsup actually did, and that's the thing that breaks the clean version of this story. He ruled that Anthropic's use of legitimately acquired books was — his word — 'quintessentially transformative.' Fair use. But retaining pirated copies of those same books? Not fair use. Same model. Same learned patterns. The only variable was how the books were obtained.

Cole Brennan: Wait — the AI doesn't know the difference, right? Like, it learned the same thing either way.

Malcolm Reeves: Identical output. The model couldn't tell you which weights came from a purchased copy versus a pirated one.

Cole Brennan: So here's what I keep — okay, imagine a student who reads every book in a library. If the library bought those books, fine. If the student snuck in and photocopied stolen copies, the law says that's a different thing. Even if what that student learned is completely identical. That's the ruling. The law is saying the permission structure matters more than what actually got transferred into the machine.

Malcolm Reeves: And the U.S. Copyright Office looked at this exact framing — the 'it's transformative because it involves computational analysis' argument — and cautioned against it. They said you can't just call AI training inherently transformative and call it done.

Cole Brennan: So the Copyright Office is pushing back on the same logic Alsup used to let Anthropic off on the legal books? That's — I mean, those two things can't both be right.

Malcolm Reeves: And that's exactly why the RIAA, NMPA, ASCAP, and BMI all filed amicus briefs in the OpenAI litigation saying training should not qualify as fair use, full stop. Irving Azoff's Music Artists Coalition joined that brief too. They're trying to close the door Alsup cracked open — before it becomes the settled standard.

Cole Brennan: So the doctrine isn't settling — it's actually fracturing. Like, Alsup draws one line, the Copyright Office draws a different one, and the RIAA is in the OpenAI case arguing neither line is in the right place.

Malcolm Reeves: Three institutions, three positions, one unresolved question. And every AI company is currently making product decisions inside that gap.

Cole Brennan: And every company making product decisions in that gap — that's the tell, right? Because the gap is also where you park the 'licensing is impossible at scale' argument. And I keep — I mean, that argument doesn't survive like a ten-second check. ASCAP and BMI license millions of compositions annually. They already did the impossible thing.

Malcolm Reeves: And GEMA didn't wait for anyone to resolve the theory. Munich Regional Court, ruled Suno was not entitled to use GEMA-represented music without a license. Full stop. That happened before American courts got anywhere close.

Cole Brennan: Wait — Germany just... ruled? And Suno kept going in the U.S.?

Malcolm Reeves: Kept going. No licensing adjustment. That's not a legal strategy — that's a bet that one jurisdiction's ruling doesn't matter to their product timeline.

Cole Brennan: So the feasibility argument is — actually, no, it's not even a feasibility argument at that point. It's economic. Like, SAG-AFTRA, the Writers Guild, the American Federation of Musicians — they filed jointly in the OpenAI litigation, framing this as economic survival for over 250,000 workers. That's not IP abstraction. That's people's livelihoods.

Malcolm Reeves: Irving Azoff's Music Artists Coalition joined that amicus coalition too — alongside CAA and the Human Consent Foundation. The labor and management sides basically landed in the same brief.

Cole Brennan: Which almost never happens. And then you put a face on it — think about a session guitarist in Nashville who's recorded 400 takes over fifteen years. Her work may be inside a model she's never heard of, generating income she will never see, and there is no mechanism for her to even find out. Warner Music Group can audit. She can't.

Malcolm Reeves: And that opacity is the thing the GEMA ruling doesn't reach — because that case is about recordings. What comes later in this story is messier: artists suing not just over their music but over their names, their voices, their musical identities. That's a different injury entirely.

Cole Brennan: Right — and you can't audit identity the way you audit a sound file. It's genuinely harder.

Malcolm Reeves: And that's exactly the wall you hit — because Jason Isbell, David Lowery, Guy Forsyth, Eduardo Calle, those four didn't sue over recordings. They alleged Suno used their names, their images, their likenesses, their musical identities. That's not copyright. That's closer to right-of-publicity law. Completely different framework.

Cole Brennan: Wait — so the injury isn't 'you copied my song,' it's 'you ingested who I am as an artist.'

Malcolm Reeves: And Reservoir's suits make the same move — it's not just the recordings, it's the artistic identities attached to Johnny Cash, Joni Mitchell, Madonna. The name carries meaning the copyright doesn't protect.

Cole Brennan: And here's what makes that — I mean, actually no, the reason that's so hard to fight is the corpus is secret. You can't audit it. Like, Isbell can't walk into court and say 'here is exactly where my identity got ingested' because nobody outside Suno knows what's in the training data.

Malcolm Reeves: The Getty versus Stability AI ruling gestured at this — England and Wales High Court, November 2025. Found against Stability AI on trademark for AI outputs. But largely favored them on copyright for the model weights themselves. Training and generation treated as two separate legal events.

Cole Brennan: So the model — the thing that holds the ingested identity — that's protected. But what it generates from that identity isn't.

Malcolm Reeves: No unified answer. And that split is precisely why identity violation and copyright violation are going to require entirely different frameworks — courts can't reach one with the tools built for the other.

Cole Brennan: Which means — cultural sovereignty, the right to know how your name and your work are being used and to decide whether to allow it, that right just... disappears the moment the corpus goes dark. OpenAI, same problem — the Times lawsuit, the RIAA amicus briefs, same opacity. You can't consent to what you can't see.

Malcolm Reeves: And that opacity point — I'm stuck on it too. Because every prior disruption had a moment where the law caught up. Player pianos, radio, YouTube — same arc every time. Technology moves without permission, creators sue, courts either carve out fair use or mandate licensing, and the industry adapts. But AI training is doing three things simultaneously: copying, learning, and commercializing. No prior framework handles all three at once. Which means if courts rule that training requires licensing, Congress would likely have to design an entirely new legal mechanism — not an expansion of copyright, not an extension of existing licensing systems. Something that doesn't exist yet.

Cole Brennan: And the lawsuits aren't waiting for that mechanism to exist. They're — I mean, that's actually what they are. They're the battering ram forcing the doctrine into existence before anyone's built the wall.

Malcolm Reeves: Which is an uneasy place to sit. The suits are doing exactly what suits do at the edge of a disruption. And yet the thing they're pointing at — a court can't build a legislative framework. That's the gap that remains when this conversation ends.

Cole Brennan: Yeah. No verdict. Just a very loud, very expensive question.

Malcolm Reeves: Good conversation. Genuinely — I'll be turning that Alsup split over for a while.