As a result of a cyberattack on the music AI service Suno, a source code leak occurred, shedding light on the true scale of data collection for training models. A hacker operating under the pseudonym ellie.191 managed to gain access to the company's systems by infecting an employee's computer with the Shai-Hulud worm. The malware spread through compromised npm packages and stole credentials, tokens, and access keys.

The leaked files contain detailed logs of audio downloads from YouTube Music, Deezer, Pond5, and several other platforms. According to my analysis, this is not just a breach of confidentiality but direct evidence of systematic piracy in the AI music generation industry.

Shocking Numbers

One of the files recorded the download of 2,013,545 music clips from YouTube Music, totaling 113,879 hours. Another set, labeled ytm_tagged, included 152,162 hours of tagged recordings from YouTube. The code also revealed 62,117 hours of music from Pond5, 12,287 hours from Deezer, 17,615 hours from the genius_hq set, and 19,514 hours from the International Music Score Library Project. Jamendo accounted for 3,726 hours, Freesound for 410 hours, and the musescore_lyrics set for 103 hours. Additionally, Suno planned to download about 1 million hours of podcasts from approximately 420,000 RSS feeds.

To collect audio from YouTube, developers used Bright Data proxy services, allowing them to route requests through third-party IP addresses, hiding the true source of the downloads.

Connection to Label Lawsuits

The leak directly echoes ongoing legal proceedings. On May 21, Universal Music Group and Sony Music Entertainment demanded that 61,026 recordings found in the training data using Audible Magic audio fingerprinting technology be added to the lawsuit against Suno. The labels claim that Suno used YT-DL and YT-DLP to bypass YouTube's encryption and convert streaming audio into downloadable files, violating Section 1201 of the DMCA. The leaked code confirms the fact of downloading from YouTube and using proxies but does not prove the circumvention of a specific technical measure.

Suno confirmed the incident, discovered in November 2025, but states that it primarily affected legacy code and that sensitive personal data was not compromised. However, the company has not published a technical report, raising questions about transparency.

My comment: This leak is not just a technical incident but a signal for the entire industry. Suno, like many other AI companies, relied on "gray" data collection methods, which now threatens it with multi-million dollar lawsuits. Investors and users should closely monitor the case's development—it could set a precedent that changes the rules of the game for all generative models.