My Music My Choice tool poisons songs to prevent AI voice cloning and deepfake songs

Artists gain protection against AI voice cloning with new tool

A wave of AI-generated music clones swept streaming platforms in 2025, with synthetic versions of tracks by Bad Bunny and Drake becoming indistinguishable from authentic recordings1

1

. The surge exposed musicians to copyright infringement and identity theft while leaving fans unable to separate real performances from deepfake songs. Researchers at Binghamton University and startup Cauth AI responded by developing My Music My Choice (MMMC), a digital safeguard that empowers artists to protect their work before it reaches the public2

2

.

The system works through audio poisoning—adding imperceptible changes to a song's waveform that human listeners cannot detect but that completely disrupt AI voice cloning models1

1

. When played on streaming services, protected tracks sound identical to master recordings. But when the same files are fed into cloning software, the microscopic alterations confuse the algorithm, causing it to interpret the protected vocals as an entirely different vocal track2

2

. Instead of replicating the artist's voice, the system produces only distorted static.

How the invisible audio technique disrupts generative AI systems

Umur Aybars Ciftci, research assistant professor at Binghamton University, and Ilke Demir, CEO of Cauth AI, designed My Music My Choice to minimize impact on human listeners while maximizing disruption for machines2

2

. The adversarial protection against vocal cloning targets the specific ways AI models process audio. "We're trying to figure out exactly which tiny modifications to introduce so that people hear no difference at all, while AI voice-cloning systems are thrown off," Ciftci explained2

2

.

Artists can apply this protection during production and release recordings with confidence that preventing AI replication of an artist's vocal track will succeed from the start1

1

. The approach differs fundamentally from existing copyright tools that attempt to catch fakes after they appear online. By poisoning recordings before distribution, musicians gain control over how generative AI systems interact with their work.

Testing reveals effectiveness across multiple music genres

The research team tested the tool on 150 music tracks spanning multiple genres, demonstrating its versatility across different musical styles2

2

. They plan to scale up testing with larger data samples and compare My Music My Choice with similar methods, though Ciftci acknowledged that few alternatives currently exist in this space2

2

.

The urgency stems from real-world consequences musicians faced throughout 2025. "People are using voice cloning for fun but also for nefarious purposes," Ciftci said, describing how bad actors grab someone's voice and make them sing things they never would1

1

. Beyond intellectual property rights violations, artists experienced emotional tolls and lost revenue as their identities were borrowed without permission2

2

.

What musicians should watch as protection scales

The research, titled "My Music My Choice: Adversarial Protection Against Vocal Cloning in Songs," was presented at the 39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop: AI for Music2

2

. Binghamton students Gerald Pena Vargas, Alicia Unterreiner and David Ponce contributed to the work2

2

.

Collaboration between academic researchers and Cauth AI provides insight into front-line industry challenges, bridging the gap between laboratory concepts and industrial-scale impact2

2

. As the team continues wider testing and refinement, the message for musicians remains clear: protection against AI cloning now arrives before the clone appears, not after copyright chaos has already unfolded1

1

. Artists should monitor developments as this digital safeguard moves toward broader availability, potentially reshaping how the music industry defends against unauthorized AI-generated music clones.