Leaked Suno source code details the platforms it scraped to train its music AI

Leaked Suno source code details which platforms the AI music generator scraped to train its models, according to code a hacker accessed and shared with reporters, published July 15, 2026. Dataset-scraping scripts and file labels point to more than two million clips pulled from YouTube Music alone.

The leak lands in the middle of a copyright fight. Record labels are suing Suno over how it trained its models, and the scraping records name specific catalogs and services rather than the broad “open internet” Suno has described in court.

The scripts and labels indicate Suno ingested 2,013,545 YouTube Music clips, about 113,879 hours of audio, dwarfing every other source. They also list roughly 12,000 hours from Deezer and 17,600 hours from Genius, plus smaller volumes from Pond5, Jamendo, the International Music Score Library Project, Freesound and MuseScore. The hacker, who used a stolen employee credential in a November 2025 supply-chain attack, also reached account data for hundreds of thousands of customers and partial Stripe payment records.

A Suno spokesperson called it “a limited security incident that was quickly contained” involving “outdated source code that is no longer in use,” and said the company did not separately notify customers of the November breach.

Suno has said in litigation that it trained on “essentially all music files of reasonable quality that are accessible on the open internet,” a fair-use claim the labels are suing to challenge. The leaked labels do not by themselves establish infringement, but they hand plaintiffs a far more specific map of what went into the training set.