Every week, more of the music arriving at distributors and labels is not made the way the people receiving it assume. Generative tools can now turn a text prompt into a finished-sounding master, and those tracks are being uploaded for distribution, pitched for sync, and folded into catalogs that change hands for real money. For the teams who sign, license and deliver music, a quiet question has moved to the front of the desk: how do you know what you are handling?
Disclosure is the honest answer, and on its own it is not enough. InterSpace Distribution made AI disclosure mandatory for every release in May, and the major streaming services have tightened their own rules through the year. But disclosure is self-reported. It depends on the person uploading the track knowing the truth and telling it, which is not a given once a file has passed through three hands before it reaches a catalog buyer. A distributor running quality control, a sync agent answering a brand’s legal team, a label weighing a catalog it wants to buy: each of them, at some point, needs a read on the audio itself.
What SEAM does
That read is what SEAM is built to give. SEAM, which stands for Sonic Evidence of AI in Music, is an audio-recognition service from The InterSpace Group that listens to a track and returns a plain, per-track verdict: Likely human, Likely AI, or Unsure. It works on the sound, not on the metadata or the uploader’s word, and it returns a likelihood alongside the verdict so a reviewer can see how strong the signal is.
The framing is deliberately modest. SEAM is an advisory signal, a read on the audio and not a guarantee. It never rejects, holds or approves a track by itself. The verdict goes to a person, and the team decides what happens next. That is the right posture for a tool like this, because the cost of treating a probability as a verdict is paid by a real artist whose own human record gets wrongly flagged.
Who actually needs this
The person who needs SEAM is rarely the listener. It is the businesses in the middle:
- Labels and A&R, screening demos and incoming catalog, who want to know what they are signing before it carries their name.
- Distributors, running quality control before a release goes to the stores, where an undisclosed AI track can mean a takedown or a strike against the account that delivered it.
- Sync and licensing teams, and music supervisors, who increasingly have to answer a direct question from a brand or a broadcaster: was any of this made by AI?
- Publishers and catalog buyers, doing due diligence, for whom a catalog’s share of undisclosed AI material is now part of what it is worth.
What these teams share is exposure. A wrong assumption about one track is an annoyance. The same wrong assumption running through a delivery pipeline or sitting inside a catalog is a liability.
How you use it
SEAM runs two ways. A team can scan a track in the web console and read the verdict there, or send tracks through the API and get the verdict back inside its own system, which is how a distributor would wire it into the delivery pipeline or a catalog tool would check a batch. The first 1,000 tracks are scanned free, so a team can test it against music it already knows the answer to before trusting it on music it does not.
What a verdict can and cannot tell you
A detection signal is only useful if you read it honestly. A few things worth holding onto:
- Likely is not proven. A Likely AI verdict is a reason to look closer and ask for disclosure, not a finding you can publish about someone.
- Unsure is a real answer. Heavily processed human production and careful AI can sit close together, and a tool that refuses to guess on those is more useful than one that always picks a side.
- It pairs with disclosure, it does not replace it. The signal is strongest as a check on what an uploader told you, not as a substitute for asking.
- The audio is the evidence. Metadata, tags and filenames can all be edited after the fact. The point of listening back is that the sound is harder to dress up.
Availability
SEAM is live now at seam.interspacemusic.com, with free scans for the first 1,000 tracks and an API for teams that want to run it at volume. It is built and operated by The InterSpace Group, the independent music company behind InterSpace Distribution and ToneGrid.
