Undisclosed AI music is a track made wholly or partly with generative AI that reaches a store without the AI declaration the store expects. For a distributor, it is a delivery problem first: the file clears intake, the metadata says nothing about AI, and the first check it meets may be the store’s own.
Stores now write AI rules assuming the delivering party already knows what is in the file. Here is what that costs, why a disclosure field alone leaves a gap, and where an audio read fits in quality control (QC).
What undisclosed AI music costs a distributor
It costs revenue, reach and standing, and the stores route all three through whoever delivered the file. Positions as of November 2026:
TIDAL says it expects distributors to flag AI content before it reaches the platform, and withholds royalties from tracks it identifies as wholly AI-generated (The Next Web). Apple Music told providers in August 2026 that AI Transparency Tags are required where AI was used to create a material portion of the content, without saying what happens if a provider fails to tag (The Next Web).
Deezer reported in July 2026 that it receives nearly 90,000 fully AI-generated tracks a day. It removes detected tracks from recommendations and editorial playlists, says it will take down those used for fraudulent streams, and lists a policy update for the companies that deliver to it among possible next steps (Deezer).
Spotify said in September 2025 that it had removed more than 75 million spammy tracks in a year, and announced a spam filter to tag uploaders as well as tracks and stop recommending them (Spotify). It also charges labels and distributors per track when it detects flagrant artificial streaming (Spotify support).
Why a disclosure field alone leaves a gap
A declaration records what the uploader says, not what the audio is. InterSpace Distribution has required AI-use declarations for cover art and audio, plus the country of recording, on new releases since May 25, 2026 (mandatory AI disclosure requirements). Any such field is only as good as the answer given.
Spotify’s AI credits appear only where artists disclose through a label or distributor, and Spotify says a missing credit does not mean AI was not used (Spotify; see also why disclosure only travels through your distributor).
An audio read gives QC a second signal that does not depend on the answer. SEAM (Sonic Evidence of AI in Music) is an audio-recognition service from The InterSpace Group. It reads the audio itself, not metadata or the uploader’s word, and returns a plain verdict (Likely human, Likely AI or Unsure) with a likelihood.
Where an audio read fits in QC
Run it after the declaration is filed and before the release is queued, on the file you intend to deliver. A mismatch found then costs a conversation; found after delivery, it means an update or takedown.
SEAM is strictly advisory. It never rejects, holds or approves a track by itself, so every verdict needs a person and a written rule. It runs in a web console and via API.
What to do with each verdict
Compare the verdict with the audio declaration.
- Likely human, no AI declared: continue normal QC alongside your other checks.
- Unsure: a reviewer listens, and the uploader is asked how the track was made. It is a question, not a flag.
- Likely AI, AI declared: the audio and declaration point the same way. Check the tool and degree are filled in, then release under your AI rules.
- Likely AI, no AI declared: the mismatch case. A reviewer decides whether to pause the release while the uploader confirms or corrects the declaration.
- Likely human, AI declared: usually consistent, since a declaration can cover AI cover art or an assistive tool on a human recording. Keep it as filed.
Keep the first message neutral. A Likely AI verdict is a reason to look closer and ask for disclosure, not proof, and not something to publish about anyone.
Our QC review flagged this release for a closer look. Please confirm whether AI tools were used in the audio and, if so, which tool and for what part of the track.
Log the verdict, likelihood, declaration and decision for every track, and use the likelihood to order the review queue. Re-run the read whenever the audio file is replaced, give uploaders a fixed window to reply, and name who can override a decision.
Because SEAM only advises, the policy that matters is yours. Write down what each verdict triggers, then run a first batch in the SEAM web console. The first 1,000 tracks are free.
