Back to blogIndustry

A Streaming Service Just Branded AI Music — and Says 60% of Its Streams Are Fake

SoundGen TeamOctober 2, 20267 min read
Share:
A Streaming Service Just Branded AI Music — and Says 60% of Its Streams Are Fake

For years, platforms treated AI music like a problem hidden in moderation queues and legal memos. Qobuz has now moved the argument into the listener's line of sight.

The hi-res streaming service has begun placing a visible label on releases it identifies as fully AI-generated. It also says 60% of streams on the AI tracks it detects are fraudulent and therefore excluded from royalty calculations.

That figure is explosive. It is also more complicated than the headline.

The label listeners can actually see

Qobuz says its new system identifies fully AI-generated music and marks it inside the app. This matters because most platform responses have happened behind the scenes: spam filters, takedowns, distributor rules and metadata standards that ordinary listeners rarely encounter.

A visible tag changes the experience. It gives listeners information before they press play, while giving platforms a category they can treat differently for discovery and payment.

Qobuz also says it has de-indexed more than four million AI-generated tracks with no listening activity. De-indexing is not the same as deleting. It means those tracks stop crowding search and recommendation surfaces while remaining in the catalog.

What “60% fraudulent” does — and does not — mean

The claim is about streams on tracks Qobuz has classified as AI-generated, not about 60% of AI musicians committing fraud. A fraudulent stream can come from a bot network, an automated replay farm or another attempt to manufacture activity. One operation can generate thousands of bad plays.

The number is also Qobuz's own measurement from its own detection and fraud systems. The company has not published a complete independent audit, and it has not disclosed enough about its classifier for outsiders to reproduce every decision.

So the responsible reading is narrow: on Qobuz, within the catalog its systems flag as AI-generated, suspicious listening is unusually concentrated. It is a warning signal, not a universal census of AI music.

Legitimate creators could get caught in the blast radius

The platform is trying to solve two separate problems at once:

  1. Origin: was the recording fully generated by AI?
  2. Behavior: were the streams produced by real listeners?

Those questions are related, but they are not interchangeable. A fully generated ambient track can have genuine fans. A human-recorded song can be pushed through a bot farm. Labeling the first does not prove fraud, and leaving the second unlabelled does not prove legitimacy.

The risk is that listeners, editors and recommendation systems begin treating an origin label like a warning label. That would punish creators who use licensed tools honestly while doing little to stop fraud attached to conventional recordings.

Why a smaller platform can move first

Qobuz is not Spotify or Apple Music. Its audience is smaller and oriented toward high-resolution listening. That makes this a leading experiment, not an industry-wide rule.

It also makes Qobuz an unusually useful test case. A focused service can change discovery and metadata faster, watch how listeners respond and adjust before the largest platforms make the same decision at global scale.

If the tag reduces spam without burying legitimate work, competitors will notice. If it becomes a stigma that produces false positives, they will notice that too.

What responsible AI musicians should do now

  1. Keep a creation record. Save prompts, stems, edits, collaborators and the model or service used.
  2. Use accurate release metadata. If a distributor asks about AI involvement, answer precisely rather than forcing a hybrid track into an all-or-nothing label.
  3. Never buy streams. A promotional service promising guaranteed plays can poison an otherwise legitimate release.
  4. Watch your traffic. Sudden plays from one location or an impossible completion pattern should be investigated quickly.
  5. Make human decisions visible. Arrangement changes, recorded parts and thoughtful credits give curators a clearer picture than a generic audio file does.

The real scarlet letter is fraud

Visible AI labels are likely to spread because listeners and rights holders are asking for disclosure. The crucial question is what platforms allow that label to mean.

If it means “this release has provenance information,” it can build trust. If it quietly comes to mean “this music is fake,” it collapses creation and fraud into one category — a mistake that helps neither artists nor listeners.

Qobuz has made the debate visible. Now the industry has to decide whether the tag becomes context or a verdict.

Sources

S

SoundGen Team

Written by the team building SoundGen

Ready to Create?

Put these insights into practice with SoundGen's AI-powered music tools.