Spotify removes 75 million spam and 'AI slop' tracks

Spotify says it has pulled 75 million spam tracks from its catalog over the past year, a figure that is hard to picture: it exceeds everything many independent labels put out in a whole decade. The company confirmed it this week and, as ABC News reports, the announcement comes with a telling shift in vocabulary. It no longer talks only about 'spam', but about 'AI slop', the term the industry uses for the flood of AI generated tracks uploaded en masse with a single goal: to scrape streams and, with them, royalties.

The problem is not AI made music in itself, but its volume and its intent. Tools like Suno or Udio can produce plausible tracks in seconds, and all it takes is automating the upload to flood the platform with thousands of near identical tracks, each one designed to capture a few cents. At the scale of millions, that noise distorts the charts, drains the royalty pool shared among real artists and degrades the recommendations that hold the whole service together.

What Spotify has put on the table

According to the published information, the measures go well beyond wiping out accounts in one sweep. They rest on three pillars:

1. Spam filter: a system that detects and stops mass uploads and automated loading patterns before they saturate the catalog.
2. AI use disclosure: Spotify relies on an industry standard, DDEX, so that credits state whether a track used AI in the vocals, the instrumentation or the production. It does not ban AI, it requires declaring it.
3. Anti impersonation rule: a tougher policy against voice clones that mimic real artists without permission, a phenomenon that blew up with the musical deepfakes of 2023.

The underlying idea, summed up in ABC's own headline, is to make AI music easier to detect rather than banning it outright.

Why it matters

For months the debate revolved around whether AI would end up 'replacing' musicians. Spotify's move points to a more immediate and less philosophical problem: streaming fraud at industrial scale. When 75 million tracks are flagged as spam, the implicit message is that the bottleneck is no longer generating music, but telling signal from noise inside a catalog of hundreds of millions of tracks.

It is also a matter of trust. If listeners stop taking for granted that a playlist contains real artists, and if labels suspect that their share of the royalty pool is being diluted among bots, the product loses value for everyone. Transparency about AI use works here both as catalog hygiene and as a commercial argument.

Who it is relevant for

The most exposed are independent artists and labels, who compete for the same royalty pot and the same slots in playlists; a filter that removes noise at scale benefits them directly. Producers who use AI legitimately will have to get used to declaring it in the metadata, something that in the medium term may become a standard requirement across digital distribution. And for anyone building products on top of music catalogs, the signal is clear: the provenance of a track, who made it and with what, becomes a first class piece of data.

At ElephantPink we have long seen that the problem with generative AI is rarely the quality of a single piece, but what happens when it is produced at industrial scale. Spotify's bet on labeling before banning strikes us as the more sensible path, though its real usefulness will depend on something still to be proven: that the filter gets it right far more often than it gets it wrong.