Sony, Universal sue AI music generators for ‘training’ models on copyrighted songs

With these lawsuits, music publishers have mounted their most direct challenge yet against AI models that are used to generate music. In 2023, Universal sued Amazon-backed Anthropic for allegedly using copyrighted lyrics to train its chatbot Claude.

What are the allegations in the lawsuits?

The lawsuits allege that Suno and Udio scraped digital sources to copy and download a massive number of sound recordings which included copyrighted recordings. It further states that the scraped recordings were included in the training datasets of the generative AI models, which form the basis of Suno and Udio’s services.

In order to confirm that the AI-generated outputs were achieved by copying and ingesting copyrighted recordings, the lawsuits state that a test was carried out wherein targeted prompts that included specific characteristics of popular songs caused the AI music generator tools to generate music files that strongly resembled the copyrighted recordings.

Additionally, the lawsuit alleged that Suno’s and Udio’s AI models had undergone “overfitting”. “ “An AI model is “overfitted” when it is too closely adapted to the data on which it was trained, making it difficult for the model to generalise to new data sets,” the lawsuits read.

The lawsuit against Suno also states that one of the AI startup’s early investors inadvertently admitted to training the model on copyrighted music when he said that he expected for Suno to be sued by copyright owners and that it was “the risk we had to underwrite when we invested in the company.”

“Rodriguez’s message was clear: he was willing to “underwrite” the costs of the lawsuits relating to Suno’s large-scale intellectual property theft because he expected his investment in Suno to be accretive despite the damages owed to copyright owners,” the lawsuit read.

The music labels alleged that they were able to prompt the AI models to generate outputs resembling popular songs by famous artists such as Mariah Carey, ABBA, Green Day, and more. In order to strengthen these allegations, the labels presented side-by-side comparisons of the musical scores of the original songs and the AI-generated ones.

In response to the lawsuit, Udio said that it was “completely uninterested in reproducing content.” “We stand behind our technology and believe that generative AI will become a mainstay of modern society,” the company said in a blog post, adding that it continues to refine its filters to avoid reproducing copyrighted works or artists’ voices.

Meanwhile, Suno CEO Mikey Shulman reportedly told Wired, “Our technology is transformative; it is designed to generate completely new outputs, not to memorise and regurgitate pre-existing content. That is why we don’t allow user prompts that reference specific artists.”

Is there a way ahead?

The music industry’s lawsuits against AI music generators does not necessarily mean that labels are against AI-generated tracks altogether as they support licenced use of copyrighted music. For instance, Universal Music Group, which is one of the authors of the lawsuits, recently announced that it was partnering with SoundLabs, a voice cloning startup.

Even artists and musicians are increasingly experimenting with generative AI tools to enhance their work. Music composer AR Rahman used generative AI to recreate the voices of dead singers for a track in an upcoming Tamil film. When questioned about his use of AI, Rahman said in a BBC interview that “any technology should benefit humanity and not take away livelihoods.”

“I personally think you should use [AI] like a slave. We should use it for the good parts and not give authority or decisions which would harm humanity, even if you think it’s 100% correct. Keep it as an augmented tool for humanity to do things which sometimes take hours and hours of work,” he opined.