Half the uploads are machines

The number that changed the conversation was not about quality. One major streaming service reported that fully AI-generated tracks passed half of its daily uploads, which turned an aesthetic argument into an operational one.

Volume alone would be survivable. A catalogue that nobody plays costs storage and nothing else, and services have carried enormous quantities of unlistened audio for years without anyone minding.

What made it urgent is that a large share of the plays attached to that material were not people. The same service reported that the overwhelming majority of streams on generated tracks it examined were fraudulent, produced by automation rather than listeners.

That is a payments problem, not a curation problem. Streaming royalties are divided from a pool, so a fake play is not a victimless inflation of a chart. It is a transfer out of everyone else's share, funded by subscribers who did not listen to it.

Platforms have moved in stages, and the stages are revealing. First detection, then labelling, then filtering out of recommendations, and only recently demonetisation, which is the step that actually removes the incentive.

One service has gone furthest, refusing to pay royalties on tracks its detection identifies as entirely machine-generated. That is a stronger position than labelling and a harder one to defend, because it requires the detector to be right.

Detection accuracy is the unglamorous centre of the whole question. A false positive demonetises a human record, a false negative funds the fraud, and neither error is visible to the artist on the receiving end of it.

The policy also has to decide what counts. A track composed by a person and mixed with machine assistance is not the same as a prompt executed at scale, and a rule written for the second will eventually be applied to the first.

Rights holders have an interest that is not quite the public one. Reducing the flow of generated material into the pool raises the per-stream value of everything else, which is a commercial motive sitting behind a stated concern about authenticity.

The honest summary is that the industry did not act when the machines started making music. It acted when the machines started collecting.