Qobuz starts rejecting AI-generated content today. FUGA is enforcing at the distributor tier.

On the week of July 13, 2026, Downtown Music Holdings' distribution arm FUGA activates rejection of 100% AI-generated content at delivery time under Qobuz's AI Charter. This is the first DSP-driven, distributor-enforced AI ban to move from "policy on a website" to "your delivery just got rejected." Here is what changes for every distributor shipping into Qobuz.

What turned on today

Two separate things aligned this week, and it matters that they are separate:

1. Qobuz's AI Charter — a public policy Qobuz has been building since 2024, committing to reject 100% AI-generated content and to compensate rightsholders in a way that does not dilute human catalogs. This is the DSP-tier rule.
2. FUGA's enforcement layer — Downtown Music Holdings (which owns FUGA, CD Baby, Songtrust and Curve) is now filtering AIGC at the distribution tier before handing tracks to Qobuz. Distributors delivering through FUGA have been notified that AIGC-flagged tracks will be rejected starting the week of July 13, with per-release notification when a track is refused. Priority releases can be escalated to FUGA support for manual review if the sub-distributor guarantees non-AI provenance.

The FUGA notice is the operationally important half. Qobuz publishing a charter is a statement of intent; a distributor with hundreds of sub-distributors turning the filter on is an actual production event. Every indie distributor that delivers into Qobuz via FUGA — which is a very long list — is now on the hook to know, at ingest, whether a track is AI-generated.

The AI Charter debate has been running for two years at the policy level. As of this week it is a returned-delivery ticket on someone's dashboard.

Why this matters more than the RIAA label proposal

Three days ago we wrote about the RIAA/IFPI proposal for AI-generated and AI-assisted labels. That proposal is important, but it is a labeling standard — you attach the correct metadata, the release ships either way. The Qobuz/FUGA move is different in one specific way:

• The RIAA proposal says: tell us what it is.
• The Qobuz Charter says: if it's 100% AI-generated, we won't take it.

"We won't take it" is a categorically bigger problem than "label it correctly." It means an indie distributor now has three failure modes at ingest, only one of which was previously live:

The new failure mode is not a fine or a strike — it is a refused delivery. That failure surfaces to the sub-distributor's operations team as a rejected release, and to the artist as a missed release date on one of the DSPs they were counting on. Nobody in the chain likes that phone call.

What FUGA told distributors, in plain terms

Reconstructing the operative pieces of the FUGA communication received by sub-distributors this week:

• Qobuz's AI Charter contractually prohibits delivery of 100% AI-generated content.
• FUGA reviewed sub-distributor catalogs and flagged deliveries over the last six months as potentially high in AIGC.
• Starting the week of July 13, 2026, FUGA's QC systems will detect and reject AIGC for Qobuz delivery, with per-release notification.
• FUGA recommends sub-distributors invest in their own AI detection capabilities and restrict 100% AI content from Qobuz delivery.
• Priority releases can be escalated to support@fuga.com for manual processing if the sub-distributor guarantees non-AI provenance.

The recommendation to sub-distributors is the operationally load-bearing sentence. It is FUGA saying, on the record, that they expect the pre-upload filter to live at the sub-distributor level, not at their level. That is the correct architecture — and it means the ~250-500 indie distributors that deliver through FUGA all now have the same problem on the same week.

What a sub-distributor needs, concretely, this week

Three capabilities, in this order:

1. Detect 100% AI-generated tracks at ingest, before generating the DDEX manifest for Qobuz. This is the immediate blocker. Not "audit later" — before delivery. The tolerance for false negatives is now measured in refused-delivery tickets and missed release dates.
2. Distinguish "100% AI-generated" from "AI-assisted." The Qobuz Charter rejects the first, not the second. Getting the distinction wrong in either direction costs money: false positives block ship-worthy human releases, false negatives get delivered and refused.
3. Produce an audit trail per decision. When a distributor escalates a priority release to support@fuga.com and says "we guarantee this is not AI," that guarantee needs to be backed by something more than confidence. Model version, score, per-module signal, timestamp, human reviewer verdict if applicable.

The audit trail requirement is the one distributors under-estimate. FUGA will not process guarantees indefinitely on trust alone. Sub-distributors that guarantee correctly build trust; sub-distributors that guarantee wrong lose escalation privileges, and eventually delivery relationships.

What DistroShield emits for the Qobuz/FUGA case

The production model (v9, deployed 2026-06-23) is a 4-class primary classifier plus an attribution stage. Mapped against the Qobuz Charter action:

Test-set metrics for v9 as deployed: human recall 98.31%, binary AI recall 98.68%, macro F1 0.92 across the four classes. In practice this means: of every 100 human-made tracks a distributor sends through, ~98 will be correctly cleared for Qobuz. Of every 100 fully-AI tracks, ~99 will be caught before the DDEX manifest goes out.

The API response already carries the three pieces needed to defend an escalation to support@fuga.com:

POST /v1/analyze
Authorization: Bearer ds_...

{
  "audio_url": "https://your-cdn/track.wav",
  "metadata": { "title": "...", "artist": "...", "isrc": "..." }
}

→ {
  "ai_score": 0.02,
  "classification": "human",
  "confidence": 0.96,
  "recommendation": "deliver",
  "model_version": "v9-4class-2026-06-23",
  "signals": {
    "primary_class_probs": { "human": 0.96, "hybrid": 0.03, "unknown_ai": 0.01, "suno_unlicensed": 0.00 },
    "fingerprint_match": null,
    "attribution": null
  },
  "analyzed_at": "2026-07-13T14:22:11Z"
}

The model_version, primary_class_probs, and analyzed_at fields are what turns a "we guarantee this is not AI" into a defensible claim. That JSON is what a sub-distributor forwards to FUGA support when escalating a priority release.

What happens next

Two predictions that are, at this point, essentially safe bets:

1. Other DSPs follow Qobuz within 6-12 months. Deezer already has AI detection running internally and demonetizes AI streams. Apple has published Transparency requirements. Spotify has anti-fraud enforcement live since 2024. Qobuz is the first to codify "we won't take it" into a distributor-facing contractual rejection — the others will import the pattern once they see it work.
2. Other upstream distributors follow FUGA within the same window. The Orchard, Believe, Ingrooves and Symphonic all have the same downstream contractual pressure from DSPs. Once one of them turns on rejection at delivery, the others do it defensively so their sub-distributors don't route around them.

The window in which "we didn't know the track was AI" is a defensible answer at a distributor operations meeting is closing. This week is the leading edge of it closing. Sub-distributors that plug in a pre-upload filter now do it once, calmly, before the ticket volume hits. Sub-distributors that wait get to do it under refused-delivery pressure with an artist on the phone.

If you deliver through FUGA

Two useful things to do this week, whether you end up using DistroShield or not:

1. Sample your last 30 days of Qobuz deliveries. Run them through any AI detector — ours, Pex, Audible Magic, a homegrown model. Any track flagged as 100% AI that you already delivered is a candidate for proactive takedown before FUGA's QC catches it and puts it on your rejection log. Quiet takedown is cheaper than public rejection.
2. Wire a pre-upload check into your ingest. One API call before your DDEX generator runs. The response gives you enough to decide deliver / hold / block per track, and enough audit trail to defend the decision upstream.

The Qobuz Charter itself is not new. What's new is that as of this week, the enforcement is on. The distributors that already had a pre-upload filter will not notice anything different next Monday. The distributors that didn't are going to have a busy week.