The AI music war becomes a data war: Suno subpoenas the MLC, the labels subpoena DistroKid

The lawsuits over AI music are no longer only about what Suno copied. They are now about numbers: how much AI music reaches streaming services, what it earns, and whether it pushes human artists down. In mid-October both sides went to court to get those numbers from companies that aren't parties to the case. Suno went after the Mechanical Licensing Collective (MLC). UMG and Sony went after DistroKid.

What is Suno asking the MLC for?

The MLC is the nonprofit, set up under the US Music Modernization Act, that has administered blanket mechanical licences for streaming and download services since January 2021, collecting royalties from those services and paying songwriters and publishers. On 15 Oct 2026 Suno filed a motion to compel the MLC in federal court in Nashville (M.D. Tenn.).

According to Digital Music News, Suno's April subpoena asked for monthly performance data for every label, broken out label by label, at Amazon, Apple, Deezer, Meta, Pandora, Snapchat, SoundCloud, Spotify, Tidal, TikTok and YouTube, from 2021 to the present. "Performance data" covers fields such as ISRC, title, label, distributor and play count. A narrower August version dropped Meta, Snapchat and TikTok. Suno's argument is that the data shows what happened before and after AI music arrived, and so goes straight at the labels' claim that AI ate into their market share and royalties.

The MLC objects. It says it doesn't collect data from labels or distributors and has no mandatory label fields. It also says the streaming services objected to disclosure. Suno's lawyer says the MLC had agreed to produce the data, with Suno paying a cost estimate of roughly $5,000, before it reversed course. Suno has also refused to limit the request to the plaintiff labels, saying that would defeat its purpose. Udio has filed a similar motion.

What do UMG and Sony want from DistroKid?

On 14 Oct, UMG and Sony asked Manhattan federal court to make DistroKid comply with a September subpoena, Billboard reports. The labels say DistroKid has refused to produce a single document. In a statement, DistroKid said it is a third party committed to transparency and will cooperate with the legal process.

The labels' proposed order is the most revealing document in the dispute. In a 9 Oct response, DistroKid described a third-party "prediction model" it uses to test uploads for AI. The labels want, month by month since DistroKid started using it:

• how many recordings DistroKid delivered to streaming services;
• how many it ran through the model;
• how many the model scored as likely AI, in whole or in part, in whatever probability bands it reports (or by decile, if it gives a continuous score);
• the streams and royalties those flagged recordings earned, and how many the model attributed to Suno.

A second request asks for the model's name and vendor, and any analysis of its accuracy, "including the basis for the statements about false positives" DistroKid made in its response.

The labels' reasoning, per Music Business Worldwide: Suno doesn't track where its outputs go once they leave its platform, they say, while DistroKid sits upstream of the streaming services. The filing says DistroKid delivered nearly 12 million tracks to one unnamed service between March and September 2026, more than half of everything that service received, as The Hollywood Reporter also reports. That is a count of all deliveries, not confirmed AI tracks.

Why do both sides want this data so badly?

Suno's main defence is fair use. One of the four factors a US court weighs is the effect of the use on the market for the original work. The US Copyright Office's 2025 report on AI training said that factor shouldn't be read narrowly, warned that the speed and scale of AI output risk diluting markets for works like those in the training data, and noted that royalty pools can be diluted too. It's a pre-publication report, not a court ruling, but it gives the labels' dilution theory a footing.

So each side is building its evidence. The labels want DistroKid's numbers to show an AI flood that is measurable and earning money. Suno wants MLC data covering the whole market to argue that any shift in label share has other causes. Warner settled with Suno in November 2025. UMG and Sony are still suing, and filed a second suit on 18 Sep.

Where are the artists in this?

Nowhere, so far. The data being fought over is built from artists' work: their plays, their uploads, the detector's verdict on their tracks. None of the filings we read gives artists a way to see it.

The commentary we found online has stayed general. A London royalty-recovery service posted on X that the DistroKid fight shows how training data and music ownership are becoming central issues in the AI era. That's true, but the more specific problem is in the proposed order itself. If DistroKid's model flags a human-made track as likely AI, that track can be counted in the labels' tally of AI streams and royalties. The artist would never know. The labels' own request for false-positive data suggests the error rate is a live question. We've written separately about how distributors decide a song is AI and what to do when they get it wrong, and about Suno's admission that v6 learned from its users' creations.

What we think

The fair-use fight will be decided in part on a detector score, and the people that score describes have never seen it.

Look at what each subpoena actually measures. The MLC exists to license and pay for songs, not recordings, and it says it doesn't collect data from labels or keep mandatory label fields. So Suno is asking for label-level performance data from a body that says it doesn't keep data in that shape. DistroKid's data is a third-party classifier's probability guesses, and the labels' own order asks how often it's wrong. Neither is a clean count of "AI music". Both will be presented as if they were.

That matters most to the artists who sit in the margins of those tables: independent musicians who use some AI, or none, and whose tracks a model has scored as likely AI. In the labels' version they're evidence of a flood. In Suno's, their streams are part of the market-wide baseline it wants to measure against. Either way, they're numbers in someone else's argument.

Two fixes would help, and neither has to wait for a verdict. First, any distributor that runs an AI detector on uploads should tell artists when a track is flagged and let them contest it, before that score ends up in court or in a takedown. Second, any settlement or licence that comes out of these cases should include public, regular reporting on how much of the royalty pool goes to AI-generated tracks, broken down so an independent artist can see it. The labels say dilution is measurable. The data exists. The people being diluted should be allowed to see it.

Sources

• Proposed order, UMG Recordings v. Kid Distro Holdings (DistroKid), 1:26-mc-510, S.D.N.Y., filed 15 Oct 2026
• Suno, Inc. v. Mechanical Licensing Collective, 3:26-mc-00021, M.D. Tenn., docket, CourtListener
• Copyright and Artificial Intelligence, Part 3: Generative AI Training, US Copyright Office, May 2025
• Our Story, The MLC
• Suno v. MLC Subpoena Dispute Boils Over Amid UMG Lawsuits, Digital Music News
• UMG & Sony Demand DistroKid Tell How Many AI Songs It Uploads From Suno, Billboard, 16 Oct 2026
• Universal Music and Sony Want to Know How Many AI-Generated Songs DistroKid Releases, The Hollywood Reporter, 16 Oct 2026
• UMG and Sony ask court to force DistroKid to hand over AI music data for Suno case, Music Business Worldwide, 19 Oct 2026
• Monetunes on X