21.2 MILLION TRACKS, AND NOW YOU CAN LOOK
The Atlantic built the thing nobody in the AI shops wanted built. Researcher Alex Reisner compiled four of the giant datasets circulating inside the AI-development community — roughly 21.2 MILLION tracks, three of the four scraped straight off Spotify and YouTube — and made them searchable. Free. No account. Type in a song title, an artist name, or an ISRC, and find out whether your record is sitting inside a training corpus. Every independent musician who suspected and couldn't prove it now has a search box. RA picked it up and the whole club-music press followed.
Being in the dataset isn't proof of infringement on its own — but it turns a vibe into evidence, and it lands while the copyright cases against Suno, Udio and Google's Lyria 3 are live. Practical: if you make music, go search your own catalogue, because the burden just shifted from 'prove they took it' to 'explain why it's in here.' This is the most consequential thing in music this week and it isn't close.
NOT A MEASUREMENT — UNCALIBRATED, UNVALIDATED
