A tenth of streams are not real

Streaming royalties are paid from a pool. A service collects subscriptions, takes its share, and divides what remains according to how many times each recording was played. Nothing about that arrangement checks whether a play was a person.

The recorded music industry is now moving on the consequence. Business Standard, citing the Financial Times, reported that the IFPI has issued guidelines under which distributors will take stronger measures to stop fraudulent music reaching streaming platforms.

The figure prompting it is the one to sit with. Industry executives estimate that close to ten per cent of music streams are now fraudulent, which in a pro-rata system is not a measurement problem but a transfer.

Because the pool is fixed, every fraudulent stream is paid for by legitimate rights holders. A fake play does not create money; it redirects a fraction of somebody else's, and at one in ten the redirection is no longer marginal.

Generative tools changed the economics of the attack rather than inventing it. Stream manipulation predates the technology, but producing a plausible catalogue of thousands of tracks used to require either musicians or obvious effort, and now requires neither.

That is why the intervention lands on distributors rather than on platforms. Distribution is the chokepoint through which an independent release reaches every major service at once, and it is the last place a human decision is made before a track enters the pool.

It is also the awkward place for the industry to apply pressure. Cheap distribution is what allowed independent artists to reach streaming services without a label, and controls at that layer fall on the people the openness was for.

The obvious response is that legitimate artists have nothing to fear from checks, which is true in principle and unreliable in practice. Automated screening produces false positives, and a false positive against an independent release means income stopped while somebody appeals.

The harder question is what a rule can even prohibit. Music generated with AI assistance is not inherently fraudulent, and a great deal of ordinary commercial production already involves it; what is fraudulent is the manufactured listening, not the manufactured song.

Conflating the two would be a mistake with a long tail, and the guidelines will be judged on whether they separate them. Policing fake streams is a fraud problem. Policing AI-assisted composition is a taste problem dressed as one.

Either way the arithmetic is what forces action. An industry can argue about authenticity indefinitely, but a tenth of the pool leaving through a hole in the floor is a number that appears in accounts.