Marketplace Tech New tools make it harder for AI to train off copyrighted music
5 snips
Aug 21, 2025 AI-generated music is on the rise, creating waves in the industry as bands like The Velvet Sundown gain traction. Many artists are worried about their work being used without permission. Enter MusicShield, a revolutionary tool that discreetly alters musical features to protect original compositions. It cleverly keeps the music enjoyable for listeners while confusing AI models. The conversation highlights the critical balance between technological advancement and artist protection in today's music landscape.
AI Snips
Chapters
Transcript
Episode notes
Viral AI Band Example
- The Velvet Sundown surged on Spotify with 70s-style rock that was entirely AI-generated.
- Jin Liu's Music Shield inserts imperceptible artifacts so humans hear the same song but AIs learn different features.
Perceptual Gap Between Humans And Machines
- Machines and humans perceive music differently, creating a perceptual gap defenders can exploit.
- Music Shield leverages that gap by injecting noises that change extracted features while remaining inaudible to humans.
Decoy Target Redirects AI Learning
- Music Shield redirects AI feature extraction toward a copyright-free target music.
- Models trained on the protected audio learn features of the target, not the original track.