Particle.news
Download on the App Store

SZA Says 238 of Her Songs Trained AI Models as Musicians Condemn Suno

The disclosures could intensify legal and industry scrutiny over how generative-music systems use recorded tracks.

Overview

  • Last week The Atlantic published a public detection tool that draws on four large training datasets and let artists check whether their recordings were included in the collections used to train music generators.
  • Over the weekend SZA posted that a search showed “music AI has trained off 238 of my songs,” urged artists not to support AI music, and accused developers of disproportionately targeting Black creators.
  • Suno responded by saying its training metadata does not include artist names, that its models are not designed to reproduce training material, and that it is improving impersonation detection.
  • Major-label actions continue to split the industry: Sony is actively suing Suno and Udio while Warner and Universal reached licensing settlements with some AI firms and the American Federation of Musicians has pursued related legal claims.
  • Beyond legal fights, artists raise practical concerns about lost earnings, cultural harm to Black and brown creators, and environmental costs from large-scale model training while Suno tests label-backed models and seeks funding.