The U.S. Copyright Office Sessions: Generative Artificial Intelligence in Music and Sound Recordings

On May 31, 2023, the U.S. Copyright Office (USCO) held the final session of its Spring 2023 AI Listening Session. This session was held across two panels and discussed the copyright implications of AI-generated content (AIGC) in music and sound recordings. The panelists consisted of various stakeholders in the music industry such as founders of AIGC music companies, songwriters, professors, and counsel to music and streaming companies.

The key issue discussed during the three-hour session was whether the training of AI models on artist’s copyrighted works, without a license, is infringement and not fair use. In light of the recent holding in Warhol Foundation v. Goldsmith, panelists considered the fair use defense and the ramifications of AIGC involvement in human generated works.

The conversation began with a discussion of the current use of AIGC in the creation of sound recordings. Views varied on whether AIGC used as a tool in a human-directed creative process should be protected by copyright law, and as to whether AI technology benefits creators and consumers. Some expressed the view that AIGC lowers barriers to entry in terms of music creation, thus making the creation of musical works more accessible. Others adopted the view that AIGC is a useful tool for complex music composition, sample design, autotune, and mastering. Still others spoke to the consumer benefits, discussing how AIGC can be used to finely tailor copyrighted music to individual tastes, e.g., Spotify’s AI DJ, which creates highly personalized playlists for users while also providing commentary about the music it is playing.

Other panelists viewed these benefits as infringement on an artist’s catalog and the training of AI models with preexisting copyrighted material. For example, panelists discussed the fact that AI singing voices are trained by timbre and can, ultimately, turn into or clone the voices that they are trained on, such that IP issues may arise when the timbre these models are trained on are from the voices of famous and well-known artists, without their permission or a license. Participants opined that, if the artist does not give consent, then it should be considered “voice-jacking.” Panelists further discussed the issue of ownership of voice, and the dangers of feeding a technology that could one day replace human created works as we know them to be.

Several participants voiced the need for clarity regarding AIGC music and sound recordings through policymaking and guidance on federal copyright law to help stakeholders navigate this new frontier in music.


The U.S. Copyright Office Sessions: “Artificial Intelligence and Copyright – Visual Arts”