The fake artist playbook, ten years later — why DSPs can't police what pads their margin

Ten years ago, Music Business Worldwide broke open the "fake artist" story at Spotify. This week, the executive who ran the largest network of pseudonymous artists ever documented on the platform joined Spotify as Managing Director for the Nordics. The story isn't a curiosity — it's the blueprint for what generative AI is doing to distributor ingest right now.

The story, compressed

Spotify's flagship mood playlists — the ones you get served when you ask for "focus" or "chill" or "peaceful piano" — were, for years, disproportionately filled with tracks credited to artists who didn't exist as real people. Much of that catalog came from Sweden-based library-music companies. Labels grumbled that Spotify was preferring this material because it came at a margin advantage: lower per-stream royalties, no publisher headaches, no artist relations.

In 2024, Swedish newspaper Dagens Nyheter identified one musician, Johan Röhr, behind a single network that dwarfed everything reported before:

The music was delivered to Spotify by Overtone Studios (previously A-P Records), which was acquired in 2022 by Epidemic Sound — the same company that had been at the centre of MBW's original 2017 reporting. When Dagens Nyheter came knocking, Overtone's CEO declined an interview but called Röhr "a pioneer in the mood music genre" in a written statement.

That same executive is now, per MBW's reporting this week, Spotify's Managing Director for the Nordics.

The structural point people keep missing

The instinct is to treat this as one bad-actor story. It isn't. The story is that the incentive gradient at DSPs points the wrong way for policing this specific category of fraud. When library-music tracks pay less per stream than major-label catalog and get injected into the mood slots users don't choose their music for, gross margin on those minutes is meaningfully higher. Policing that supply would cost the platform money.

This is the structural version of a conflict of interest, and it's older than generative AI. You do not need to allege a single specific villain to see it: any large marketplace that both hosts and curates supply, and earns a different margin depending on which supply gets shown, has a bias toward the higher-margin supply. Detection of "is this real, is this fake, is this what it claims to be" gets underprioritized because doing it well would hurt the P&L.

Why this matters now

Generative AI didn't invent catalog gaming. It industrialized it. The Röhr network needed a decade to build 656 fictional artists. Someone with a laptop, a few AI music tools, and an ingest account at a permissive distributor can build the same footprint in a weekend. The bottleneck stopped being production and started being ingestion — which is exactly where the fake-artist economics push against enforcement.

What changed in 2026 is that the DSPs themselves started drawing lines: Spotify pulled 75M+ AI tracks in twelve months, Deezer reports over half of daily uploads are now AI-generated and executes retroactive takedowns, Qobuz and FUGA block unlicensed AI at ingest. But every one of those enforcement actions happens after a distributor has already accepted the track and pushed it downstream. The economic exposure — chargebacks, retroactive royalty clawbacks, license breach, EU AI Act Article 50 fines starting next year — lands on the distributor, not the DSP.

Where detection has to live

If ten years of the fake-artist story teach one operational lesson, it's this: waiting for the DSP to catch it means the exposure has already crossed onto your ledger. The layer that can act early — before ingest, before delivery to stores, before payouts start flowing — is the distributor. Every track that comes through your ingest pipeline needs three things answered before it ships:

• Is the audio human, AI, or hybrid — and if AI, is it licensed AI (Suno's new partner model, BMG-licensed generation) or unlicensed?
• Does this recording already exist somewhere else under a different artist name, different distributor, or different ISRC — the modern equivalent of "is this a pseudonym network"?
• Is there a signed, timestamped verdict on file for every accepted track — so that if a DSP takes it down retroactively, you have the forensic record to defend the roster and appeal?

None of those three questions is answerable by asking the DSP. They aren't structurally motivated to answer them for you, and by the time they act you're already past the point of clean recovery.

The playbook didn't die. The tooling has to catch up.

Ten years of pseudonymous mood-music injection tells you that policing supply from the demand side is a losing position. The lesson generalizes: whoever accepts the track first is where the truth has to be established. That's the ingest layer, not the store layer. That's the distributor.