SZA, Kenneth Blume Decry Use of Their Songs in AI Training Data Sets
Get the Culture newsletter
Daily culture — film, music, books, the trends and ideas worth your attention. Free.
- The Atlantic debuted a detection tool built by researcher Alex Reisner that indexes four training data sets encompassing over 21 million songs, pulling from catalogs ranging from Bad Bunny, Taylor Swift, and Beyoncé to independent artists.
- SZA discovered 238 of her songs in the training data and wrote on Instagram that AI companies disproportionately exploit Black artists: "I AINT HEARD A WHITE AI SONG YET.. why so disproportionate?"
- Producer Kenneth Blume (FKA Kenny Beats) singled out Suno, writing: "I can't imagine being proud to earn a paycheck obliterating the work and dreams of artists."
- DJ Sabrina the Teenage DJ found 22 of her songs in Suno's data set and flipped the narrative on listeners who had accused her music of sounding like "AI slop" — the AI had literally learned from her.
- Three of the four data sets led to songs via YouTube and Spotify links, and Reisner noted that developers often scrape those links with automated tools that can bypass logins and ad revenue, potentially violating both platforms' terms of service.
- Hudson Mohawke offered a contrarian take, writing on Instagram that the entertainment and tech industries have never been "fair" or "moral" and that illegal sampling has long been a cost of doing business.
- Suno and Udio have both been sued by major labels; Warner Music signed a licensing deal with Suno last year, and the American Federation of Musicians sued UMG and Warner earlier this month over AI use of members' music.
Why it matters: The tool gives artists concrete evidence — 238 songs for SZA, 22 for DJ Sabrina — of an unregulated 21-million-song training pipeline that major labels are now fighting in court while simultaneously licensing some of the same companies. Independent and Black artists, who SZA says lack legislative protection in creative and medical fields, are the most exposed, and the YouTube/Spotify scraping method raises platform-liability questions neither the labels nor the streamers have resolved.
Ask SkimNews
