Suno AI: millions of songs from YouTube and Deezer, not exactly borrowed

A recent incident has unveiled the secret behind Suno's musical prowess: millions of songs and lyrics were scraped from online platforms. Data obtained from a hacking incident exposed the sources used to train the AI music generator.

The news, reported by 404 Media and then The Verge, indicates Suno extensively pulled content from YouTube Music, Deezer, and Genius. This wasn't a "borrowing" with permission, but a massive scraping operation. It's like a musician learning by copying every single note without asking.

Suno has consistently kept its training datasets under wraps. They always stated they used "a mix of data" without further specifics. We now know this "mix" included, without authorization, copyrighted material from YouTube Music, Deezer, and Genius.

Suno's secrecy about its training data now appears to have a very clear reason. Revealing they had taken millions of tracks and lyrics from unlicensed platforms would have sparked a legal uproar from day one.

AI companies often sidestep discussions about their training data. The reason is simple: most models are trained on vast amounts of web content, much of which is copyrighted. Suno, with its millions of "collected" tracks, is another confirmation of this questionable practice.

This incident raises serious questions about the future of AI-generated music. If models are fed with others' works without compensation, what will be the impact on original artists? It's somewhat like building a castle with stolen bricks and then claiming it's original.