A landmark incident has occurred in the generative music industry. A hacker under the pseudonym ellie.191 managed to infiltrate the systems of the popular AI service Suno using the Shai-Hulud worm. The malware, spread through compromised npm packages, stole an employee's credentials, granting access to the source code and internal logs. The result is a public disclosure of exactly how Suno collected data to train its models.
Analysis of the stolen code reveals impressive download volumes. One file recorded the acquisition of 2,013,545 music clips from YouTube Music, equivalent to 113,879 hours of audio. An additional dataset labeled ytm_tagged contained another 152,162 hours of tagged recordings from the same platform. Besides YouTube, the dataset includes 62,117 hours from Pond5, 12,287 hours from Deezer, 17,615 hours from Genius HQ, and 19,514 hours from the International Music Score Library Project. To bypass YouTube's restrictions, Suno developers used Bright Data proxy services, masking requests as traffic from various IP addresses.
Podcast Plan and Scale of the Leak
Of particular interest is a plan to download about 1 million hours of podcasts from approximately 420,000 RSS feeds. Whether this was actually implemented is unclear from the published data. The hacker himself claims to have accessed information on hundreds of thousands of users, including emails, phone numbers, and data related to the Stripe payment gateway. Suno has confirmed the breach, dated November 2025, but insists the incident was quickly contained and did not affect sensitive personal data, and that the leaked code is outdated. However, the company has yet to provide a technical report that would allow independent verification of these claims.
Legal Battles and a New Twist in the Dispute
This leak comes amid fierce legal battles with major music labels. Universal Music Group and Sony Music Entertainment have already demanded the addition of over 61,000 recordings to their lawsuit against Suno, which they claim were used for training without a license. The labels allege that Suno violated the DMCA by circumventing YouTube's technical protection measures using tools like YT-DLP. The leaked code, demonstrating the audio download process and proxy usage, could become significant evidence in this dispute, although it does not in itself prove the circumvention of encryption.
My analysis: This incident is not just another leak. It is a direct demonstration that many AI companies, in the race for data, are crossing all conceivable and inconceivable boundaries. Suno, while attempting to legitimize its activities through a partnership with Warner Music, simultaneously stole content from competitors and independent artists. The outcome of this case could set a precedent that forever changes the rules of the game for the entire generative AI industry.