Generative AI has upended the traditional definition of how a musician's work is "used," sparking debate over whether compensation is owed for training data or only for final outputs. Companies like Sureel and SoundVerse are now building systems to bridge that gap. These initiatives seek to restore the economic principle that creators earn more as their work is utilized more frequently.
Sureel, a startup recently acquired by Warner Music Group, has partnered with the Swedish copyright agency STIM to explore royalty frameworks for AI training data. The effort comes amid accusations that the generative AI industry represents "the biggest act of copyright theft in history." Proponents argue that musicians deserve ongoing compensation when their creative essence lives on inside a model's structure.
The technical challenge lies in tracking which musical works contributed to a model's outputs and calculating fair payment per use. Unlike streaming or radio, where plays are easily counted, AI models encode training data in complex, non-transparent ways. SoundVerse and others are developing attribution tools to make this accounting possible.
If successful, these systems could transform the relationship between artists and AI companies from adversarial to collaborative. Artists would gain a new revenue stream, while AI firms would secure a legal and ethical path to train on copyrighted material. Major labels and copyright agencies are watching closely, with Sweden's STIM serving as an early testbed.
Skeptics question whether any tracking system can truly capture the diffuse influence of training data on model outputs. The risk of unresolved disputes could still stall adoption across the industry.