Qobuz Just Cut Off AI Music's Royalties. Spotify Still Hasn't.

Qobuz built its own detector for AI-generated music, and it just published the receipts. Deputy CEO Georges Fornay says the system has already identified more than 4 million AI tracks that got zero listener engagement, pulled straight from search. Of the streams the system does flag as AI, 60% turn out to be fraudulent, engineered to farm royalties rather than reach an actual listener, and those get cut out of payment reporting entirely.

That is a genuinely hard line. Most platforms still treat AI content as a grey zone: label it, maybe, but keep paying out on the stream count. Qobuz is saying no, if the fraud detection catches it, the money stops.

Is this actually a hard problem for Qobuz to solve?

Here is the number that undercuts the victory lap: AI-identified tracks are 0.38% of Qobuz's total streams. That is a rounding error. Qobuz is a paid, hi-res, audiophile-first platform with a small, deliberate catalog and a subscriber base that skews toward people who already research what they buy. Catching AI floods there is closer to weeding a garden than draining a swamp.

Compare that to Deezer, which started tagging AI-generated tracks before Qobuz did. Deezer disclosed back in mid-2025 that AI-generated content made up roughly 18% of everything being uploaded to its platform, a completely different order of magnitude, on a completely different metric (uploads, not streams, so the two numbers are not directly comparable, but the scale gap is the point). Spotify, with its open upload pipeline and hundreds of millions of tracks, has talked about tackling AI slop and spam but has not published anything close to Qobuz's fraud-and-royalty-exclusion numbers.

"Musical discovery must remain guided by human choices. Everyone should be able to know what they are listening to."

That is Fornay's framing, and Qobuz backs it with a real editorial firewall: flagged content cannot appear in Qobuzissimes, Albums of the Week, human-curated playlists or the algorithmic recommendation engine. Removal from the catalog altogether, Qobuz says, is reserved for confirmed fraud, not for AI use as such. The company draws a specific line between AI as a production tool (mixing, mastering, assistance a human artist uses) and "industrial production of AI-generated content, without human creative intervention, designed to saturate catalogs or defraud royalty systems." One gets tolerated. The other gets defunded.

Why does the industry care what a small hi-res platform does?

Because the economic case is not niche. Qobuz's own AI Charter, published in February 2026, leans on CISAC-derived projections that AI-generated content could cost human creators as much as 10 billion euros in lost revenue by 2028, a potential 24% hit to creator income, while the AI systems themselves could be pulling in roughly 4 billion euros a year from unlicensed use of that same human-made work. Whether or not those exact figures hold up, they are the numbers labels and collecting societies are already citing in royalty and licensing fights, and Qobuz just built a working enforcement model around them, catalog scanning, fraud-rate disclosure, and payout exclusion, that a bigger platform could copy if it wanted to.

What we think

Qobuz deserves credit for publishing numbers most platforms would rather bury, especially the 60% fraud rate among flagged tracks, which is the closest thing to hard evidence yet that AI-stream farming is a real royalty-theft problem, not a hypothetical one. But the honest read is that Qobuz solved the version of this problem it was always best positioned to solve. The test that matters is whether Spotify, Deezer or Apple Music will publish an equivalent fraud rate and cut equivalent payouts once AI content stops being 0.38% of their traffic and starts being double digits. Right now, only Deezer is even in the tagging conversation, and its 18% AI-upload figure suggests the flood is already a lot closer to the front door.

Why it matters

If detection-and-exclude becomes the industry benchmark, every label and every artist gets a concrete number to point to in royalty disputes. If it stays a boutique-platform talking point, the AI-stream fraud problem keeps compounding on exactly the services that move the most money.