An investigation by The Atlantic has revealed that millions of copyrighted songs have been used to train AI music models, including tracks from popular artists like Taylor Swift and Bad Bunny. The publication created four searchable databases that collectively encompass 12 million, 9 million, and two additional databases with approximately 100,000 songs each.
The article by staff writer Alex Reisner provides insight into the extent of copyrighted music included in AI training data. Legal actions are currently underway against generative AI music platforms such as Suno and Udio, which assert fair use as a defense for using copyright-protected material. A previous lawsuit in the book publishing sector struggled to advance on copyright claims, while piracy allegations gained more traction. The initial settlement from the book publishing case amounted to $1.5 billion, with final outcomes and payouts still pending.
The databases from The Atlantic may serve as valuable resources for the music industry in pursuing future lawsuits related to copyright infringement. In response to the rise of AI-generated music, many streaming services have implemented measures to prevent, identify, or label such creations. However, the effectiveness of these measures has varied. Additionally, scammers have exploited the situation by creating imitations of established bands to capitalize on their work through AI-generated copies.
