Not all creative works face equal risk in AI’s copyright wars. Lyrics stand apart: brief, catchy, and instantly recognizable, they’re uniquely vulnerable to copying and uniquely easy to prove in court.
The risk came into focus in the Universal Music Group lawsuit against Anthropic. A federal court gave Anthropic an early win by refusing to halt its training, but the fair use question remains unresolved. The case underscores a key vulnerability: unlike a 300-page novel, a three-minute pop song can be copied—and proven copied—far more easily, a distinction that could shape the future of AI and creative ownership.
We spoke with Brian McBrearty, a Forensic Musicologist at Musicologize and Founder of royalty-free production music library Sonitarium, whose work sits at the epicenter of this collision. With a background providing infringement consulting and similarity analysis for clients including Google and Facebook, McBrearty offers a perspective grounded in the science of sound and the art of creation. His focus is not on legal strategy, but on the forensic evidence of what comes out of these powerful new tools.
- A different currency: "Lyrics are a different currency because they're short-form relative to a book. When you're writing lyrics, short phrases tend to be catchy," McBrearty explained. "Lyrics are far more likely not just to infringe, but to be recognized or observed as similar. So the risk is somewhat higher for lyrics."
While the legal war is being waged over the ethics of training AI on copyrighted data, McBrearty argued the more practical battleground is focused on the output. He sympathized with the idea that AI "training" is akin to a human musician transcribing songs to learn, but his professional work begins when that learning process produced something that sounds too familiar.
-
To what degree: "That's really where my bright line is," he said. "If ordinary prompting yields recognizable protected expression, that's very much where someone like me would come in and evaluate whether it's copying of protected expression, and whether it's enough, and to what degree, and how much of it is too similar to something else."
The author is going to bear the responsibility. They are the gatekeeper for whether their new song is original or not, and whether it is going to sound too much like something else. That's going to be burdensome on the future creators who are going to be inclined to use AI-based tools, especially if the outcome of these cases leaves doubt as to how often the tool is going to spit out derivative material.Brian McBrearty - Forensic Musicologist | MusicologizeThis focus on outputs over inputs creates a profound shift in responsibility. As AI tools become more integrated into the creative workflow, the burden of ensuring originality falls squarely on the shoulders of the human creator. McBrearty suggested that the artist becomes the final checkpoint against infringement.
-
The creator's burden: "The author is going to bear the responsibility. They are the gatekeeper for whether their new song is original or not, and whether it is going to sound too much like something else," he stated. "That's going to be burdensome on the future creators who are going to be inclined to use AI-based tools, especially if the outcome of these cases leaves doubt as to how often the tool is going to spit out derivative material."
The new reality has precedent. McBrearty pointed to MIDI’s arrival in the 1980s, when computers first linked with synthesizers and “put all kinds of power in the hands of everybody, for better or for worse.” Just as MIDI blurred the lines between human and machine-made music, today’s AI tools blur the lines between inspiration and infringement. "Technological revolutions come along, and at the time they might look one way," he reflected, "but as time passes, they begin to look a different way." The future of AI and music won’t be decided by technology alone, but by how creators, courts, and audiences choose to define originality in a world where copying has never been easier.