Michael Smith invented one thousand and forty listeners

Michael Smith, fifty-four, of North Carolina, made hundreds of thousands of AI-generated songs and one thousand and forty accounts to listen to them. The accounts worked across four platforms — Spotify, Apple Music, Amazon Music and YouTube Music — and were built to behave like people.

The arithmetic is the most interesting part. Each account played roughly 636 songs a day. Together that came to 661,440 streams every day, worth around $3,307. Close to a hundred thousand dollars a month; more than a million a year.

Spreading it thin was the whole trick. Had that many plays landed on a handful of songs, the platforms would have noticed, so the traffic was smeared in a fine layer across hundreds of thousands of tracks. Which tells you precisely what a single stream is worth: to earn one salary, he needed one thousand and forty imaginary people listening without pause.

The point is less what he earned than where it came from. Royalties are divided out of a shared pot, so every fake play shaves a little from what reaches everyone real. The prosecutor, Jay Clayton, put it briefly: “Although the songs and listeners were fake, the millions of dollars Smith stole was real.”

On 19 March, Smith pleaded guilty to conspiracy to commit wire fraud and agreed to forfeit $8,091,843.64. He is sentenced before Judge John G. Koeltl in Manhattan on 29 July, and faces up to five years. The cents at the end of that figure may be the most eloquent detail in the case: a sum is only that exact when a machine was doing the counting all along.