Musicians Confront AI Training Data Sets—But Proof Remains Elusive

Musicians discover their work in AI training databases

When The Atlantic's AI Watchdog project released a search tool allowing musicians to check whether their songs appeared in four massive digital datasets, the response was immediate and visceral. Artists ranging from independent creators to major stars like SZA discovered their work listed in collections including the Free Music Archive—databases originally created for noncommercial purposes but potentially used by AI music companies like Suno and Udio to train their models.

The revelation struck a nerve in a community already grappling with decades of declining compensation from streaming and digital distribution. But according to Charles Alexander, a Nashville-based songwriter, digital strategist, and lecturer in MTSU's recording industry program, the reality behind those search results is more nuanced than many artists initially understood.

The gap between inclusion and exploitation

Alexander, who co-founded a company focused on protecting artistic work through digital fingerprinting, emphasized a critical distinction: appearing in a dataset does not automatically mean a song was used to train a specific AI model.

"Just because your music was found in these data sets, it didn't necessarily mean that those songs and those data sets were used to train AI models at the generative AI audio companies," Alexander explained. While companies like Suno and Udio have acknowledged using publicly available data for training, they haven't specified which datasets they actually employed.

The challenge for artists seeking legal recourse is proving their work was used. "If your music was used to train a commercial AI product, you have grounds to go after those people from a copyright standpoint," Alexander noted. "But you also need to be able to prove that that event occurred. And the proving part is where I think all this kind of craters."

Legal action and legislative responses

Days after The Atlantic's investigation, a law firm known for class action victories joined an existing lawsuit against Suno and Udio brought by Delgado Entertainment Law on behalf of independent musicians. Alexander welcomed the effort, noting that independent artists have lacked protection mechanisms available to major label artists.

Meanwhile, legislative efforts are advancing. Tennessee passed the Elvis Act in 2024, and the federal No Fakes Act—which addresses name, image, and likeness protections—is moving forward. Alexander advocates for a bundle of legislation, including the proposed TRAIN Act, which would specifically target unauthorized use of training data.

Why it matters

This controversy represents a turning point where AI's impact on creative industries becomes tangible and personal for working artists. Unlike previous technology disruptions in music, AI training raises fundamental questions about consent and compensation that existing copyright frameworks struggle to address. The difficulty of proving which datasets trained which models creates a legal gray area that could define how AI companies operate—or are constrained—for years to come.

The broader implications

Alexander's central message to artists: any content posted online is vulnerable to exploitation. He recommends reading terms of service agreements carefully and understanding what permissions are being granted.

Looking ahead, Alexander believes public awareness will shift as AI impacts become more personal. "We're gonna be about six months to a year away from them caring," he said, predicting AI will affect elections, shopping, and other daily activities in ways that force broader societal engagement with these issues.

These details were first reported by WPLN, Nashville's NPR station, in an interview conducted by music journalist Jewly Hight.