The Atlantic has created and published a searchable database documenting music datasets used to train AI models, containing records of approximately 21 million tracks across four major datasets. The database enables musicians and rights holders to identify their work in training sets, addressing long-standing opacity around AI model training data provenance. Two of the datasets are particularly large, containing 12 million and 9 million tracks respectively.

What This Means for Your Business

Music and entertainment companies should use this database to audit their catalog's inclusion in training datasets and understand potential licensing implications. For AI developers, the database's existence signals increased pressure to document training data sources and implement transparent licensing practices. Organizations building generative AI systems should expect similar transparency demands across other domains.