Three major record labels said today they have filed lawsuits against the generative artificial intelligence music startups Suno Inc. and Uncharted Labs Inc., better known as Udio.

The two separate lawsuits against Suno and Udio allege that the AI companies have carried out “widespread infringement” of copyrighted sound recordings at an “almost unimaginable scale.”

Leading the lawsuits is the Recording Industry Association of America, which represents plaintiffs including Sony Music Entertainment, Warner Music Group Inc. and Universal Music Group N.V. They argue that Suno and Udio have illegally copied famous songs and tracks created by their human artists to train their AI models to create music that will eventually “saturate the market” with machine-generated content. They believe that AI-generated music will ultimately compete with their own music, and perhaps one day overshadow the genuine, human-produced originals.

“Building and operating [these services] requires at the outset copying and ingesting massive amounts of data to ‘train’ a software ‘model’ to generate outputs,” the plaintiffs said in the lawsuit. “For [these services], this process involved copying decades worth of the world’s most popular sound recordings and then ingesting those copies [to] generate outputs that imitate the qualities of genuine human sound recordings.”

Lawyers for the music labels argue that Udio has “flouted the rights of copyright owners in the music industry” since the day it was launched. It added that the startup did this in a “mad dash to become the dominant AI music generation service.”

“Neither Udio nor any other generative AI company can be allowed to advance towards this goal by trampling the rights of copyright owners,” the lawyers said.

The record labels are said to be seeking an injunction that would bar Suno and Udio from continuing to train their AI models on copyrighted songs. They also want to be paid damages from the infringements that have already been made.

Udio did not respond to a request for comment, but Suno Chief Executive Mikey Shulman told media in a statement that his company’s technology is “designed to generate completely new outputs, not to memorize and regurgitate pre-existing content.”

He added that the company doesn’t allow users to generate music using prompts that reference specific artists. “We would have been happy to explain this to the corporate record labels that filed this lawsuit, but instead of entertaining a good faith discussion, they’ve reverted to their old lawyer-led playbook,” Shulman said.

Sounds familiar

Suno and Udio have emerged as two of the most prominent companies in the nascent generative AI music industry, thanks to the way they can generate instrumentals, lyrics and vocals together in the click of a button. Udio is best known for creating what many consider to be the first real generative AI-generated music “hit” in the shape of “BBL Drizzy,” which was generated by the comedian King Willonius and has racked up millions of views on YouTube. Meanwhile, Suno launched in December and hit the headlines last month when it closed on a $125 million funding round led by prominent Silicon Valley investors such as Lightspeed Venture Partners and Matrix.

Neither company has admitted that they use unlicensed copyrights as part of their training datasets, though Udio’s co-founder David Ding previously told Billboard that his company’s models were trained on “good music.”

But there are signs that both startups have indeed used copyrighted music as part of their training data. In an article published in Music Business Worldwide, Ed Newton-Rex, the founder of Fairly Trained, a startup that certifies AI companies that don’t use any copyrighted work without a license, showed how he was able to generate music on both platforms that bore a “striking resemblance” to copyrighted music.

In the lawsuit, lawyers for the music labels also cited a number of AI-generated tracks as circumstantial evidence that Suno and Udio are using copyrighted content to train their models. They include a song generated by Suno titled “Deep down in Louisiana close to New Orle” that closely resembles the lyrics and style of Chuck Berry’s “Johnny B. Goode,” and a second track called “Prancing Queen” that sounds suspiciously similar to the ABBA hit “Dancing Queen.”

Moreover, some of Suno’s backers appear to have admitted that the company is using copyrighted music. In an interview with the Rolling Stone, Suno investor Antonio Rodriguez conceded that the startup does not have licenses for the music it was trained on. He added that this isn’t a concern for him, explaining that the threat of a lawsuit was “the risk we had to underwrite when we invested in the company.”

“If we had deals with labels when this company got started, I probably wouldn’t have invested in it,” Rodriguez said. “I think they needed to make this product without the constraints.”

Many AI companies have tried to argue that training their models on copyrighted materials falls within the doctrine of “fair use,” which allows for some protected works to be reused without breaking the law. While fair use has traditionally been used to justify things like news reporting and parody, AI companies such as OpenAI argue that it applies equally to them, because they insist that the content their models generate is still entirely new.

But the lawyers for the music labels appear to have anticipated such an argument. In their lawsuit, they stated that the companies “cannot avoid liability for their willful copyright infringement by claiming fair use. The doctrine of fair use promotes human expression by permitting the unlicensed use of copyrighted works in certain, limited circumstances, but [the services] offer imitative machine-generated music — not human creativity or expression.”

Sharing the spoils

Constellation Research Inc. analyst Holger Mueller said the lawsuits against Suno and Udio could well be a landmark case for the AI industry, and perhaps help to define where the line falls between AI-generated content that’s considered to be genuinely new content and that which is merely a rehash of copyrighted content.

“This case will have repercussions in areas such as image generation, text creation and more,” Mueller predicted. “It may also pave the way for AI companies and music labels to start making content licensing deals, similar to what we have already seen in the news industry.”

Ultimately, some kind of collaboration between AI companies and music producers is likely, said Rob Enderle of the Enderle Group. He pointed out that human creators are actually similar to AI in the fact that both learn from the authors that came before them. Whereas AI models are trained on data, humans simply gather knowledge from existing content over the duration of their lives, he said.

“If you are too close to the source material, you will face copyright infringement exposure, no matter if you are human or AI,” Enderle said.

The analyst proposed a solution in the form of an independent AI system that can review AI-generated songs and rank them based on their likeness to existing songs. Any songs that fall have an unacceptable risk of copyright infringement can then be deleted, he said. This would enable AI music generation services to adopt a two-tier subscription model that allows users to pay a small fee to create unique songs they can share within the service, and a much bigger price for creating music that they want to make money from.

“Some of the proceeds could go to the artists who created the original source material as compensation for the use of their intellectual property,” he said. “The mechanisms for this largely exist today between artists that want to license existing music and either record it with their voices or create something even more unique. This process now needs to be automated so it can be applied to AI services at scale.”

There are reasons to think the music industry would be open to such a model. RIAA Chief Executive Mitch Glazier pointed out that many record labels have already embraced “responsible developers” in the AI industry that are building sustainable tools centered on human creativity that put artists and songwriters in charge. “We can only succeed if developers are willing to work together with us,” he added. “Unlicensed services like Suno and Udio that claim it’s ‘fair’ to copy an artist’s life’s work and exploit it for their own profit without consent or pay set back the promise of genuinely innovative AI for us all.”

The RIAA’s lawsuit isn’t the first time the music industry has gone after AI companies they believe are using copyrighted material. UMG was previously among several plaintiffs that sued the AI company Anthropic PBC for distributing copyrighted song lyrics generated by its Claude 2 chatbot.

Although many people believe AI-generated music won’t quite replace human creators completely, there are justifiable fears in the industry that AI content could limit their ability to make money from the content they create.

Image: SiliconANGLE/Microsoft Designer