Suno’s Latest AI Model Got Some Label Support. But Not All of Them

Earlier this month, AI-generated music platform Suno released a new model that it made with cooperation from major record labels in an effort to smooth out claims that the company was illegally using copyrighted material to train its model. Unfortunately for Suno, the labels it worked with on this model did not include Sony and Universal, and those companies claim songs from their artists are still in the training data. The companies are now suing the startup, according to a report from Variety.

Sony and Universal’s primary contention is that, despite the fact that Suno trained its “v6” model in part on material that it got permission from other labels to use, it also trained it on outputs from its previous models. Those models allegedly were trained on material that Suno didn’t have permission to use, meaning that the model was still effectively trained on the stolen data.

“Training a ‘new’ model on the outputs of an infringing model does not eliminate the infringement; it launders it,” the labels wrote in their complaint, filed in the U.S. District Court in the District of Massachusetts. “V6 is not a fresh start; it is the fruit of the same poisoned tree.” (Notably, some very AI-like phrasing there.)

Suno, as one might expect, rejects the allegations made in the lawsuit. A spokesperson for the company told Gizmodo that the claims are “fundamentally flawed on both the facts and the law.” The company also offered an explanation for why it can seemingly produce music that accurately evokes a particular artist whether or not it specifically trained its model on that artist’s material.

“When someone asks for a particular artist, we translate that into the musical qualities behind the request – not so that a user can imitate someone else’s work, but so that people learn the language of music and make something of their own,” the spokesperson said. They also explained that its v6 model was “trained on content licensed from our partners, interactions including creations and preference signals from our community, and the accumulated learnings from our team.”

Suno notably got hacked earlier this year, which revealed some of the material that was seemingly used to train previous versions of its model. Included in the breach were what appeared to be more than two million music clips scraped from YouTube Music, more than 17,000 hours worth of songs from Genius, and thousands of hours’ worth of material from other platforms.