Hack Reveals Suno Trained Its AI on Music Scraped from YouTube, Deezer and Genius

The outlet 404 Media has published a report based on data from Suno's hacked source code — Suno being a popular tool for generating music with artificial intelligence.

According to the journalists, the code obtained reveals the mechanism by which the company spent years collecting audio material from public sources to train its model.

What the hack revealed

Based on the materials that 404 Media obtained, Suno downloaded music and podcasts from platforms such as YouTube, Deezer and Genius. What is at issue is decades of accumulated content, which was used as training data for the generative model.

The outlet stresses that it was precisely the hacked source code that made it possible to pin down the specific sources from which the data was gathered — the company had not previously disclosed any details about the origins of its training corpus.

Context

Suno is one of the best-known AI-based music generation services, allowing users to create tracks from a text description. Tools like these have repeatedly become the subject of controversy over the question of whether it is lawful to use someone else's content to train neural networks without the consent of the rights holders.

The 404 Media publication adds new details to the ongoing debate around the data-collection practices of companies developing generative AI, and it may bear on further claims from the music industry and from the platforms whose content ended up being used without explicit permission.

Source: 404 Media