Suno Admits It Used YouTube Audio to Train Its AI Music Models

Suno has acknowledged in Massachusetts federal court filings that it used audio obtained from YouTube to train its AI music models. The admission gives the copyright fight brought by major record labels a concrete new focus: not only what music may have entered the training data, but how it was collected.

The filings were submitted September 1 in the U.S. District Court for the District of Massachusetts. Universal Music Group, Sony Music and Capitol Records are suing Suno, alleging the company used licensed music without authorization to train its models.

The route to the training data

The labels allege that Suno used tools including YT-DL and YT-DLP, software designed to download audio from online video, to extract copyrighted recordings from YouTube. They say the tools were used to bypass YouTube’s technical protections. Those are allegations by the plaintiffs, not findings in the case.

Suno frames the case as fair use

Suno’s lawyers argue that using copyrighted music qualifies as fair use because its models make a new product that ultimately does not infringe copyright. They also accuse the three labels of using anticompetitive tactics to extend an unlawful monopoly over music production and commercialization.

The dispute has produced different outcomes

• GEMA, a German licensing agency, previously won a copyright-infringement case against Suno over alleged training on songs from its repertoire without licenses or songwriter compensation.
• Warner previously settled its lawsuit against Suno through a licensing deal intended to bring a licensed model into Suno’s offerings.

A narrower question inside a wider music fight

The Warner agreement shows that licensing can reshape a dispute between an AI music company and a rights holder. The Massachusetts case, however, is testing a different question: whether Suno’s asserted fair-use defense can answer the labels’ claims about copyrighted recordings and the route through YouTube.