Technology

Can You Tell If a Song Was Made by AI? The Music Industry Wants a Label

Martin HollowayPublished 10h ago3 min readBased on 5 sources
Reading level
Can You Tell If a Song Was Made by AI? The Music Industry Wants a Label
Image by itay-verchik from Pixabay

Mitch Glazier, chairman and CEO of the Recording Industry Association of America (RIAA), discusses AI-generated fakes in a new episode of The Verge's Decoder podcast, published on 9 November 2026. The conversation covers how AI deepfakes can be detected and labeled, and what role platforms such as Spotify and YouTube play.

A deepfake is a realistic fake made by software. In music, it can mean a new song that copies a singer's voice, even though the singer never recorded it.

Glazier's history with the RIAA goes back to the Napster era. He worked at the association as a lawyer during the peer-to-peer lawsuits, when music was shared illegally between users. That is one reason he often compares AI to Napster. In a separate Billboard video interview, titled "RIAA CEO Mitch Glazier on AI, Copyright & Congress", he discussed AI transparency and standardized labeling, and whether AI is the next Napster.

The RIAA says it works to protect artists' creative freedom and intellectual property. It runs a Gold & Platinum Program, which the RIAA says defines success in the recorded music industry. It also offers enforcement resources to help people tell legal music activity from illegal piracy. Michele Ballantyne is the RIAA's chief operating officer.

In March 2022, Glazier and David Israelite, president and CEO of the NMPA (National Music Publishers' Association), released statements supporting the Tillis/Leahy proposal. The proposal encourages private companies to work against online piracy.

The broader context here is that these statements point to a steady preference for voluntary, industry-led steps, backed by the law. Labeling and detection fit the same pattern. A label added at upload, or checked when a song is distributed, is a data problem as much as a legal one. It touches the systems platforms already use to match recordings and manage rights.

The original Napster dispute was about copies of recordings passing between users without a license. Generative AI raises a different question. Its output can copy a performer's voice or style without copying any existing file. Catching a copied file is like matching a photo against a known picture. Catching an imitation is closer to judging whether a signature is genuine. The Decoder episode focuses on how such signals could be recognized consistently across services.

In my view, the most useful open question is interoperability, meaning whether different services can read and respect the same label. A labeling scheme that works on one streaming service but disappears at the next helps neither listeners nor artists. Standardized labeling, which Glazier raised in the Billboard interview, only matters if platforms, distributors and rights holders agree on what a label means and how it stays attached to a file.

There is also a more hopeful reading of this moment. A clear label lets listeners choose music made by people, or music that discloses AI assistance. It also gives artists a documented basis for objecting when a synthetic copy of their voice appears under their name. Whether that becomes working infrastructure will depend on the platforms as much as on the labels.