Jul 2024

AI Music-Generators Sued for Copyright Infringement


In what is becoming this decade’s version of the Napster streaming lawsuits from the early 2000s, major music publishing companies have sued two startups for allegedly using copyrighted music without a license to “train” their artificial intelligence music generators. The two defendants both offer online services that purport to use AI to create new music based on user prompts.

The Parties to the Lawsuits

Two lawsuits were filed on the same day in two different federal courts; one was filed in the Southern District of New York against Uncharted Labs, Inc., the company behind the Udio AI music service. The second complaint was filed in the District of Massachusetts against Suno, Inc. The company recently announced it had raised $125 million to “accelerate product development and grow our team of music makers, music lovers, and technologists” for its AI music service. Suno launched the “Suno for Mobile” app on the iTunes store days after the lawsuit was filed.

According to the two complaints, the plaintiffs are a group of 13 record companies and recorded music businesses that “together, own or exclusively control copyrights in a great majority of the most commercially valuable sound recordings in the world.” They include well-known labels such as Capitol Records, Sony Music Entertainment, Atlantic Recording Company, and others.

What the Lawsuits Are About

Notably, the lawsuits don’t claim that Udio or Suno shouldn’t be able to use AI to generate music, even music that may resemble that created by the human artists whose copyrights the plaintiff recording labels control. Their focus is first on how the two companies’ services “learned” to make music – what information was used by the services to train the AI tool at the heart of each company’s music-generating services.

The music companies contend that the defendants should have secured licenses for their respective AI tools to access and use the labels’ music as part of their training. Their complaints argue that neither defendant disclosed precisely how its music-generating AI service was trained and, in the absence of licenses to either company, the “reasonable inference” is that the defendants copied the plaintiffs’ music without a license. Hence, the copyright infringement claims.

The music companies base their complaints on more than this reasonable inference, however. Their second argument is that the way the two AI services respond to “targeted prompts” – user input that the AI tool uses as guidance for generating music in response – also shows that the defendants are using the plaintiffs’ copyrighted content to generate music that sounds similar to the music companies’ originals.

Each complaint alleges that the plaintiffs tested the defendant’s system by “using a series of prompts that pinpoint a particular sound recording by referencing specific subject matter, genre, artist, instruments, vocal style, and the like.” The complaints claim that in response to those specific prompts, the defendant’s music-generating system “repeatedly generated outputs that closely matched the targeted copyrighted sound recording.” According to the plaintiffs, this means that the defendant “copied those sound recordings to include in its training data.”

Each complaint provides examples of copyrighted recordings that the plaintiffs argue were closely copied by the defendants’ AI systems based on specific targeted prompts. Some of the recordings listed in the complaints include:

For Udio: My Girl (The Temptations); American Idiot (Green Day); Sway (Michael Bublé); All I Want for Christmas is You (Mariah Carey); Billie Jean (Michael Jackson); I Get Around (Beach Boys); Dancing Queen (ABBA)

For Suno: Johnny B. Goode (Chuck Berry); Rock Around the Clock (Bill Haley & His Comets); I Got You (I Feel Good) (James Brown); Great Balls of Fire (Jerry Lee Lewis); The Thrill is Gone (B.B. King); Sway (Michael Bublé)

The Defendants Respond to the Complaints

As of this writing, neither Udio nor Suno have formally responded in court to the music companies’ complaints. Udio issued a statement one day after the  lawsuit was filed that didn’t directly reference the lawsuit but claimed to “offer some insight into how our technology works.” It then described how AI models “learn from examples” and admitted that “our model has ‘listened’ to and learned from a large collection of recorded music.” Udio argued that “we are completely uninterested in reproducing content in our training set” and that the company had implemented “state-of-the-art filters to ensure our model does not reproduce copyrighted works or artists’ voices.”

That said, Udio’s statement appears to confirm that its system does “listen to” – or use – existing music. While Udio did not clarify what existing music its system was trained on, it’s very possible (if not extremely likely) that some of the music it was trained on included recordings whose copyrights are owned or controlled by the plaintiffs. The other question raised by Udio’s statement relates to its “state-of-the-art filters.” If those filters are designed to ensure the Udio music generator doesn’t reproduce “copyrighted works or artists’ voices,” is Udio’s tool using those copyrighted originals or information derived from those originals to filter out that kind of content?

Suno didn’t issue a formal statement, but its CEO was quoted as saying that its “technology is transformative; it is designed to generate completely new outputs, not to memorize and regurgitate pre-existing content. That is why we don’t allow user prompts that reference specific artists.”

This again doesn’t directly address the music companies’ arguments that Suno used copyrighted recordings without a license to train its AI, and that by using targeted prompts, it is possible to have the Suno AI “create” music outputs that “resemble specific recording artists and specific copyrighted recordings.”

Next Steps

These cases are in their infancy, as is the use of AI for generating creative content. These two cases (and others, such as the Anthropic case I discussed in November of 2023, which, as of this writing, is continuing) may very well set the standard for whether or how AI companies must have a license to be able to use copyrighted musical recordings and compositions as training media for their AI systems.

For Your Copyright Questions, Contact Carrie Ward

Carrie Ward is an actual human (and not an AI-generated bot) with years of experience in intellectual property law, including copyright. She’s worked with media companies, filmmakers, musicians, and podcasters, helping them protect and defend their work. With a background that includes working as an in-house attorney at ABC and the company now known as Audacy, Carrie has a client-based focus that helps her understand legal issues from the client’s perspective.

Carrie also has experience with advertising and promotions law, has prosecuted thousands of FCC applications of various types, and regularly serves as outside general counsel for her corporate clients.