A hacker who breached AI music startup Suno has revealed that the company scraped millions of songs from YouTube Music, Deezer, Genius, and other sources to train its models, according to data shared with 404 Media. The breach also exposed customer information for hundreds of thousands of users, including emails and Stripe payment details. The disclosure provides one of the most detailed looks yet at how a major AI music generator built its training dataset.
What the Hacked Data Shows
The attacker, who goes by ellie.191, accessed Suno's source code from 2023 and 2024 that contained scraping instructions. One file listed targets including "genius_hq, youtube_music, freesound, jamendo, imp, deezer, ytm_tagged," with notes that non-music would be filtered out. A file called "youtube_music" reported having ingested over 2 million music clips. Another file detailed datasets totaling hundreds of thousands of hours: 113,879 hours from YouTube Music, 62,117 from Pond5, 19,514 from IMSLP, 17,615 from Genius, 12,287 from Deezer, and smaller amounts from Freesound, Jamendo, and Musescore. Combined, the datasets represent decades of music.
The code also showed Suno used proxies from scraping company Bright Data to pull songs from YouTube, and that it searched specifically for acapella versions to extract vocals. Additionally, Suno used PodcastIndex to identify 420,000 podcasts with at least five 30-minute episodes, aiming to download roughly 1 million hours of podcast audio.
Legal Context and Company Response
The RIAA has sued Suno for copyright infringement, alleging it stream-ripped songs from YouTube and circumvented technological protections. Suno has admitted in court filings that its training data included "essentially all music files of reasonable quality that are accessible on the open internet," arguing that its use of copyrighted works is fair use. The hacked data corroborates the RIAA's accusations about YouTube scraping.
A Suno spokesperson confirmed a "limited security incident" in November 2025, but said it involved outdated source code and no sensitive personal information was compromised. The spokesperson reiterated that Suno's models were trained on publicly available music files and metadata. Suno also stated it does not have access to customers' full credit card numbers in Stripe, and that individual breach notifications were not required under applicable privacy laws.
The hacker told 404 Media they breached Suno by compromising an employee via a supply chain attack, and provided a sample of customer data that included emails and phone numbers. Some affected customers confirmed they were never notified of the breach. The hacker said they had no specific motivation, stating "I like to hack anything and everything."
Industry Implications
The Suno breach adds to a growing list of revelations about AI companies scraping copyrighted content from the internet. Nvidia and Runway ML have previously been shown to scrape YouTube en masse. While many AI firms now openly acknowledge training on copyrighted material, they continue to defend the practice as fair use. Suno's spokesperson emphasized that the company designs its models to prevent replication of existing artists' work, noting that artist names are intentionally excluded from training metadata to encourage original creation.