For the first time, a US federal appeals court has ruled on whether training an AI system on copyrighted work counts as fair use — and the answer was no. The case isn't about music at all; it's about legal research. But the record industry and music publishers lined up behind the winning side, and every song on this site is made with an AI collaborator, so we've been reading it closely. Here's what was actually decided, what wasn't, and where we think it leaves people making music the way we do.
The Short Version
On 29 September 2026, the US Court of Appeals for the Third Circuit upheld a February 2025 ruling that Ross Intelligence infringed Thomson Reuters' copyright. Ross, a legal-tech startup that shut down in 2021 under the weight of this lawsuit, had used material derived from the headnotes in Westlaw — Thomson Reuters' short summaries of the legal points in court decisions — to train its own AI legal-search tool. In an opinion written by Judge Tamika Montgomery-Reeves, the appeals court agreed that Ross couldn't shelter behind fair use, and ordered it to cover Thomson Reuters' costs of the appeal.
Why Fair Use Failed Here
Fair use is the part of US copyright law that lets you use someone's work without permission in some circumstances — commentary, parody, research, or making something genuinely new. Courts weigh several factors, and two did most of the work here. The first is whether the new use is "transformative" — whether it adds a new purpose rather than just repackaging the original. The original judge, Stephanos Bibas, found it wasn't, because Ross took the headnotes to make it easier to build a rival to Westlaw itself. The second is harm to the market for the original. A tool built from Westlaw's own work to replace Westlaw is about as direct a substitute as you can get.
That's the key idea to hold on to: substitution. The problem wasn't that a machine learned from copyrighted material in the abstract. It was that the learning went into a product designed to compete with the very thing it learned from.
Why the Music Industry Cared
In November 2025, the Recording Industry Association of America and the National Music Publishers' Association filed a joint brief in the case. Their argument was blunt: training an AI model on copyrighted works so it can compete with and substitute for those works can never be fair use. They pointed out this was the first chance for a federal appeals court to answer that question, and they brought music into it directly — citing Deezer's figures that roughly 90,000 fully AI-generated tracks were arriving on its platform every day by mid-2026, more than half of uploads on peak days, and arguing that this flood dilutes the royalty pool and pulls money away from the human artists whose work was used without permission. They also noted that licensing deals between labels and AI companies now exist that didn't when the case began. Disney, Paramount, Sony Pictures, Universal and Warner Bros. filed their own brief on the same side.
What It Doesn't Decide
This is the bit worth being careful about. Ross's tool was not generative AI — it found and returned existing legal opinions; it didn't write new ones. Judge Bibas was explicit in his original ruling that "only non-generative AI" was in front of him. So this is not a ruling on Suno, Udio, or any other music generator, and it doesn't override other decisions that have gone the other way: in 2025, a different federal judge found that training a general-purpose generative model on books was fair use, in a case brought by authors against Anthropic. Fair use is decided case by case, and a tool that makes new songs is a different animal from one that hands you back the original.
The music cases are very much still live. Suno has settled with Warner, BMG, Believe and TuneCore, but Universal and Sony are pursuing it — including the second lawsuit over v6 we wrote about last month — with summary judgment motions not due until mid-2027. Udio has licensed most of the majors while still arguing in court that its copying is fair use. What the Ross ruling gives the labels is a strong appellate precedent for the substitution argument, and you can be sure it'll be quoted back at the music generators.
Where We Stand
We're not lawyers, and nothing here is legal advice. But we do have a view, because this project only makes sense if we're honest about how the music gets made.
First, we think the substitution test is a fair one, and we'd rather be on the right side of it than argue our way around it. Petunia and I isn't trying to replace anybody's records. Every track is a human idea, shaped, chosen and edited by a person, made with an AI collaborator who's credited by name — not an invented band, not a stream of unattended uploads designed to soak up royalties. The kind of AI music the RIAA and NMPA were warning about — volume for its own sake, made to compete with the artists it learned from — isn't something we want to exist either.
Second, we don't think the honest response is to pretend the legal questions don't apply to us. The tool we make our music with is in court over how it was built. That's exactly why we welcomed Suno moving toward licensed training data with v6, and why we think licensing deals are the right direction for everyone — including the songwriters, whose share of that new market is still being argued over. Licensed, traceable training is the future we'd choose, even if it costs more to get there.
Third, a word of caution about over-reading this. Some of the coverage frames the ruling as the end of the road for AI music; it isn't. It's one strong signal from one court about one kind of system. Generative tools that genuinely produce something new may still be judged differently. What we'd take from it is simpler: where your sound comes from, and how honestly you describe what you put out, matter more every month. Provenance is turning into part of the product.
So we'll keep doing what this project was built on — crediting both halves of the duo plainly, being open about the tools, and following these cases as they move. When one of the music cases actually gets decided, we'll write it up here.
Source reporting: Music Business Worldwide, with additional background from Reuters via Claims Journal and Copyright Lately.
