As the use of generative artificial intelligence continues to expand into creative industries, the music world is entering a particularly thorny chapter. The ability of AI models to generate new songs from a few user prompts has opened up new creative possibilities, but also triggered serious legal questions about copyright ownership and infringement, fair use, and the definitions of what qualifies as artistic expression.

The concern is not hypothetical. In 2023, several major record labels filed federal lawsuits against two AI music platforms, Suno and Udio, alleging these platforms copied vast catalogs of copyrighted sound recordings without permission in order to train their generative systems. According to the complaints, the AI models are capable of producing outputs that closely resemble well-known recordings, raising concerns about both the legality of the training process and the potential market impact of the resulting songs.

The layered nature of music copyright complicates the issues. Unlike books or photographs, which are often governed by a single copyright, a piece of music typically involves at least two: one in the musical composition, and one in the sound recording of the actual performance. The copyright law also provides for compulsory licensing of cover versions of songs, which further muddies the waters.

Defendants in these cases argue that their models generate new music rather than copies, and that training on copyrighted material is an “intermediate use” protected under the fair use doctrine. They also point out that many songs used as examples in the complaints are public domain compositions or works for which the plaintiffs do not own all relevant rights (as noted, a biproduct of the complex nature of music copyrights).

The record labels claim that AI-generated “synthetic” music threatens to displace human artistry and devalue the original works used in training. Whether courts will view the use of copyrighted recordings to train generative models as infringement or fair use remains to be seen.  While the products resulting from the use of the copyrighted works seem significantly transformative, the market effect (often considered the most important factor in the fair use analysis) is arguably significant.  The record labels claim that AI-generated songs are displacing human-made music and could flood streaming platforms, thereby reducing the value of the original recordings used in the training of the AI systems. The owners of the systems dispute this, saying they are used for niche and personal uses with no effect on the market for the original works.

While the courts are grappling with numerous questions regarding the legality of various aspects of gen-AI and the use of copyrighted works as training data, music presents uniquely challenging issues that will likely provide evolving interpretations and significant precedent in the coming decisions.