Legal and economic disputes over AI-generated songs are escalating as platforms capable of creating lyrics, melodies, vocals, arrangements and sound effects within minutes become more widespread. The battle is focused on ownership of the resulting works, rights to the material used to train models, and the impact of the flood of machine-generated tracks on artists’ earnings.
Works produced entirely by artificial intelligence are not eligible for copyright protection in the United States.
Protection for human elements depends on whether the creative contribution to writing, composing, modifying or arranging the work is sufficient.
Rights holders and record companies object to models being trained using copyrighted works without a licence or financial compensation.
The flow of machine-generated songs could increase competition for streams and reduce human creators’ share of streaming revenue.
Human contribution determines the scope of protection
In the United States, the U.S. Copyright Office concluded that works produced entirely by artificial intelligence are not eligible for copyright protection, while elements involving sufficient human creative contribution may be protected, such as adding original components or creatively modifying, selecting and arranging outputs.
The office explained that writing text commands known as “Prompts” is not, on its own, sufficient under current circumstances to establish human authorship. Accordingly, a user does not necessarily become the rights holder of an entire song simply because they asked a generation platform to produce a work with specific characteristics.
The chances of protecting human-created elements increase as the user’s role in writing the lyrics, composing the melody, or arranging and modifying the song expands, making the assessment of rights dependent on the details of each work and the extent of the human creative contribution to it.
Training on copyrighted works before the courts
The dispute is not limited to ownership of outputs; it also extends to the recordings and works from which models learn. Music-generation tools require vast amounts of data, prompting objections from rights holders and record companies to the use of copyrighted works in training without a licence or financial compensation.
In Germany, the music-rights organization GEMA, which represents music rights holders, filed a lawsuit against Suno, accusing it of using copyrighted recordings to train its model. On July 31, a court in Munich ruled that the company had infringed copyright by using protected songs without a licence, and ordered it to disclose profits linked to the use and pay compensation whose amount had not been determined at the time. Suno said it was considering appealing the ruling.
In the United States, Sony Music Entertainment escalated its dispute with the Udio platform, filing a new lawsuit accusing the startup of infringing the copyright of musical works.
These confrontations indicate that music companies are seeking to make works used to train models licensed, paid-for content, rather than allowing AI companies to use them without financial agreements with rights holders.
Millions of tracks and growing competition for revenue
The phenomenon is affecting artists’ earnings on streaming platforms, amid the possibility that millions of machine-generated songs will flood the market and increase competition for streams. The U.S. Copyright Office cited estimates of about 170 million AI-generated music tracks, warning that this abundance could reduce human creators’ share of revenue.
The battle thus revolves around three interconnected questions: Who owns the resulting song? What rights do the owners of works used to train the algorithms have? And who receives the revenue when huge numbers of machine-generated songs are released? As the tools continue to evolve faster than the legal rules, the amount of human creativity required to protect each work remains subject to the details of each case and the assessment of the relevant authorities.
