MMMC Tool Stops AI Vocal Cloning for Musicians

“It Shuts Them Down”: New MMMC Tool Fights Back Against Unauthorized AI Vocal Cloning

Researchers at Binghamton University and Cauth AI launch ‘My Music My Choice’ to protect artists’ voices from being hijacked by generative models.

Marcus J. Delany

Mar 4, 2026, 2:54 PM PT · 2 min read

Professor Zhiyuan Yan and Cauth AI developed MMMC to protect artists from vocal deepfakes.

THE RUNDOWN

Binghamton University and Cauth AI launched the My Music My Choice protection tool.
The technology adds invisible audio modifications that break AI vocal cloning models.
MMMC forces generative AI to output distorted noise instead of a clean voice.
Professor Zhiyuan Yan led the research to defend artists from unauthorized deepfakes.
The tool targets the technology used in viral tracks like Heart on My Sleeve.

The era of the unauthorized deepfake song may finally be coming to an end. Researchers at Binghamton University and the startup Cauth AI have officially launched My Music My Choice (MMMC), a defensive tool designed to safeguard an artist's vocal identity. The technology aims to prevent generative AI models from creating high-quality voice clones without the creator's permission.

Developed by a team led by Research Assistant Professor Umur Aybars Ciftci and key collaborator Ilke Demir, the CEO and founder of Cauth AI, the MMMC tool works by applying "adversarial perturbations" to a song's audio waveform. These modifications are completely imperceptible to human listeners, ensuring the music sounds perfect to fans. However, the changes act as a digital poison for AI software. When a generative model attempts to process the protected audio, the internal modifications cause the output to collapse into distorted noise rather than a clean vocal replica.

The launch comes as the music industry continues to grapple with the fallout from viral AI tracks. In 2023, the song Heart on My Sleeve—which used AI to mimic Drake and The Weeknd—racked up millions of plays before being pulled from streaming services. Professor Ciftci and his colleagues at Binghamton University designed MMMC to give power back to the creators, ensuring their unique vocal signatures cannot be harvested for training data without a license.