Suno AI’s Music Generator Trained on Scraped Content from Major Streaming Platforms
Hacked files from Suno AI, obtained by 404 Media, reveal the company scraped millions of songs and lyrics from platforms like YouTube Music, Deezer, and Genius to train its AI models.
Hacked data from Suno AI has provided a rare look into the company's training practices, confirming that its AI music generator was built by scraping millions of songs and lyrics from various online audio platforms. According to The Verge, which cites reporting from 404 Media, the leaked files include source code from 2023 and 2024 alongside specific instructions for pulling audio from YouTube Music, Deezer, Genius, Pond5, Jamendo, Freesound, and the International Music Score Library Project. The leaked materials suggest Suno utilized a third-party service, Bright Data, to scrape YouTube, even searching for a cappella versions to isolate vocal audio.
The scale of this data collection is significant. Files obtained by 404 Media indicate that at the time of the last update, Suno had consumed 2,013,545 clips from YouTube Music. Further documentation shows the company compiled datasets containing hundreds of thousands of hours of content from YouTube Music, alongside thousands of hours from Deezer, Genius, IMSLP, Jamendo, and Pond5. Additionally, the code suggests Suno sought to download approximately one million hours of podcasts through the tool PodcastIndex, while also scraping hundreds of hours of lyrics from Freesound and MuseScore.
This revelation arrives amid ongoing legal battles regarding Suno's training methods. The Recording Industry Association of America (RIAA) has filed a lawsuit alleging that Suno used copyrighted materials to train its models. According to The Verge, the RIAA filed an amendment last year claiming Suno intentionally circumvented YouTube’s copyright protections through stream ripping. In response to these allegations, a Suno spokesperson told 404 Media that the company's models are trained on publicly available music files and related metadata accessible on third-party websites on the open internet, arguing this constitutes fair use.
The breach also exposed customer information, including email addresses, phone numbers, and Stripe payment details. While Suno confirmed to 404 Media that it became aware of a security incident in November 2025, the company maintained that the situation was contained. A spokesperson stated that the incident primarily involved outdated source code no longer in use and claimed that no sensitive personal information was compromised. Suno further asserted that it does not have access to full credit card numbers via Stripe and determined that individual notifications were not required under applicable privacy laws.
The leaked files offer a concrete look at the data acquisition strategies that have largely remained opaque. Because Suno has historically avoided disclosing the contents of its training datasets, the 404 Media report provides a rare glimpse into the company's reliance on platforms like YouTube and Deezer. As the legal system continues to weigh the RIAA's claims against Suno's fair use defense, these documents highlight the tension between the rapid development of generative AI and the copyright protections governing the creative works used to fuel these models.