Symphonic adds MusicShield, a tool that poisons your track against AI training

Symphonic has partnered with ArtyShield to give its clients tools built to keep their music and voices out of AI systems without permission. The deal, announced August 6, 2026, puts ArtyShield’s technology inside the Client Offerings section of SymphonicMS through single sign-on. At the center is MusicShield, which protects a recording before it goes out.

MusicShield modifies what ArtyShield calls the “machine-perceived acoustic features” of a track. The edits are inaudible to people, and the point is to make the recording harder for an AI system to interpret, learn from, or train on, while the listening experience stays exactly as it was.

“

MusicShield was built around a simple idea: artists deserve meaningful protection before their work becomes part of AI systems.

How MusicShield’s inaudible noise degrades an AI model’s training data

The technique is not new to researchers. MusicShield builds on HarmonyCloak, a 2024 tool developed by a team led by Jian Liu during his time as a professor at the University of Tennessee, Knoxville, working with PhD student Syed Irfan Ali Meerza and Lehigh University’s Lichao Sun. The work was presented at the IEEE Symposium on Security and Privacy.

The underlying idea is to embed noise that tricks a generative model into treating a track as if there is nothing in it worth learning, so music trained on the protected file comes out degraded. It is the audio version of Glaze and Nightshade, the tools visual artists have used to disrupt AI image generators. Most protection products in music work after the damage: they detect a clone, flag a scrape, or file a takedown. MusicShield runs before release, which is the part that makes it different.

Symphonic CEO Jorge Brea framed the timing directly. “The conversation around AI shouldn’t begin after someone’s work has already been copied or misused,” he said, adding that opportunities in AI “have to be built on consent, transparency, and respect for the people creating the music in the first place.” The tie-up follows Symphonic’s work with Sureel AI, now owned by Warner Music Group, and a separate SourceAudio partnership offering an opt-in licensed marketplace for AI training data.

Where MusicShield stops working, and what it cannot undo

Protection that works by fooling a model is a moving target, and ArtyShield’s own release notes show it. The current build, MusicShield MS-Core v2.3.1, exists to improve “audio quality for sparse and light music with adaptive local masking, reducing audible noise in quiet passages while preserving protection effectiveness.” Read that plainly: on quiet, sparse material the protection was audible enough to need fixing, and quality and protection strength pull against each other.

The bigger limit is scope. MusicShield only covers what you run through it from now on. Anything already released, already scraped, and already sitting in a training set is untouched, and the catalogs documented inside existing AI training databases are not going to be unlearned because you protected your next single. Hacked source code from Suno listed YouTube Music, Deezer, and Genius among the platforms it scraped, per Music Business Worldwide.