How Should a Distributor Record AI Flags? What UMG and Sony Asked DistroKid
Distributors that screen uploads for AI-generated music often work from a detection model's score; DistroKid's comes from a third party. The request UMG and Sony filed against DistroKid in October shows what that score can turn into once it leaves the dashboard: a record that someone else can ask to see.
As of November 5, 2026, the court has not ruled, and DistroKid's response is due November 6. The list of what the labels asked for is still a practical checklist for any distributor or platform that flags AI uploads.
What did UMG and Sony ask DistroKid for?
UMG Recordings, Capitol Records and Sony Music Entertainment filed a motion on October 14, 2026 in federal court in New York to compel DistroKid to produce documents under a third-party subpoena. According to Music Business Worldwide, they want the data as evidence in their 2024 copyright lawsuit against Suno, which is pending in Massachusetts.
The proposed order asks for monthly figures going back to when DistroKid began using a third-party "prediction model" to test tracks for AI generation, and a further request covers the model itself. Together they ask for:
how many recordings DistroKid delivered to streaming services, and how many it tested
how many the model scored as "likely AI-generated, whether in whole or in part"
the total streams and royalties those flagged recordings earned
how many the model attributed to Suno as the actual or likely source
the name and vendor of the prediction model, and the basis for statements about false positives that DistroKid made in an October 9 response
On October 27, both sides signed a filing saying they "are working on a resolution" and extended DistroKid's deadline to November 6. The filing does not say what a resolution would involve, or whether any data will be handed over. DistroKid has said it is a third party in the litigation and will "continue to cooperate with the legal process."
Why would a detection log matter in a copyright case?
Suno argues in the Massachusetts case that training its models on copyrighted recordings is fair use. As Music Business Worldwide notes, a court weighing that defense considers whether the use harms the market for the original works.
Our reading: counts of flagged tracks and the royalties they earned speak to that market question. That would turn a distributor's internal screening data into evidence about scale, which is a use the people who set the model's threshold may not have planned for. The court has not said whether it agrees that the data is relevant.
What should a distributor record for each AI flag?
We mapped each item in the request to a record a distributor or platform could keep. This is our checklist, not a legal standard.
What the labels asked for | Record to keep | Why it matters |
|---|---|---|
Name and vendor of the model | Model name, vendor, version and the date each version went live | A score means little without knowing which model produced it |
Tracks delivered, tested and flagged per month | Monthly counts, including tracks that were never tested | Shows coverage, not just hits |
"Likely AI-generated, whether in whole or in part" | The threshold and label definitions, and when they changed | "Partly AI" can describe very different tracks |
Streams and royalties of flagged tracks | What happened after each flag: delivered, labeled, held or removed | Separates a detection flag from an enforcement decision |
Basis for false-positive statements | How the error rate was measured, appeals received and their outcomes | Public claims about errors can be asked to show their evidence |
How do other services describe their AI flags?
Services that talk about AI music publicly describe three different things, and it helps not to mix them up.
Detection. Deezer says it tags AI-generated music at the album level "to reduce the risk of incorrect classifications" and removes detected tracks from algorithmic and editorial recommendations. It also publishes its own error claim: fewer than one genuine human-made track in every 10,000 flagged incorrectly. That figure comes from Deezer, not an independent test.
Disclosure. Spotify supports AI credits submitted through labels and distributors using the DDEX standard. Its own update notes that "the absence of a credit doesn't mean AI wasn't used."
Policy. Distributor rules change. Rolling Stone reported that Believe, which owns TuneCore, had blocked Suno creations from reaching streaming services for months before it partnered with Suno in September 2026. A flag recorded in March may have been handled under a different policy than one recorded today, so the policy version belongs in the record too.
Where does a detection tool like MusicScanner fit?
MusicScanner, MuseBlossom's AI music detection tool, has been in open beta since June 2026. It returns a signal on how likely a track is to be AI-generated, which a review team can use to decide what to look at more closely.
The questions in the table apply to MusicScanner as much as to any other vendor: which version scored the track, which threshold you used, and what you did next. A detection score estimates how a track was made. It does not decide whether a track infringes anyone's rights.
What doesn't this case tell you yet?
The court has not ruled on the request, and a resolution could mean that no data is produced at all. We do not know which model DistroKid uses or how its error rates were measured. This is a US discovery dispute, and rules on producing third-party data differ elsewhere.
FAQ
Does this case require distributors to detect AI-generated uploads?
No. DistroKid is a third party in the Suno lawsuit, and the request concerns data it already had from its own screening.
Does a "likely AI-generated" flag mean a track infringes copyright?
No. A detection score is an estimate of how a track was made. Infringement is a separate legal question that depends on what was copied and how.
What happens next?
DistroKid's response is due November 6, 2026, unless the parties settle the request first. The court has not yet ruled on the labels' proposed order.
Evaluating AI music detection for an upload pipeline? Ask us about MusicScanner.