The thing the AI music industry feared most has finally happened.

On July 15, a hacking incident pushed Suno into the spotlight. A hacker gained access to Suno's source code system through an employee's credentials and discovered a shocking truth: Suno's AI music model was trained on millions of songs scraped from platforms like YouTube, Genius, and Deezer.

404 Media first broke the story. Internal files leaked by the hacker show Suno's training dataset contains decades of music — from classic oldies to the latest pop hits, from indie tracks to chart-toppers, covering almost everything. And Suno never obtained authorization from copyright holders for any of it.

The reason this matters isn't just that Suno is a leader in AI music — it's that this is yet another concentrated eruption of the copyright problem in AI-generated content.

Why Is Everyone "Scraping in Secret"?

Let's be honest: Suno isn't the first to do this, and it won't be the last.

AI music, AI art, AI writing, AI video — nearly all generative AI models are trained on scraped data. The only difference is: some companies got caught, some didn't; some got sued, some didn't.

Why does everyone scrape in secret? Because proper licensing is too expensive. If you want to train an AI music model and properly license from the three major labels (Universal, Sony, Warner), the licensing fees could be billions of dollars — and that's before accounting for the time cost of negotiating with countless independent artists. Most startups simply can't afford it.

Why don't rights holders proactively license to AI companies? Because they haven't figured out pricing yet. License too cheap and they worry AI will cannibalize their own business. License too expensive and AI companies can't afford it. Both sides are locked in a stalemate — AI companies scrape in secret, rights holders look the other way, waiting for the right moment to settle scores later.

Suno getting hacked is like poking a hole through this paper wall. Before, it was an "open secret." Now it's "public evidence." If rights holders want to sue, they've got the evidence ready-made.

Is There a Solution to AI Content's Copyright Dilemma?

The outcome of this will probably be Suno paying damages and settling. But the deeper question remains: how should copyright for AI-generated content actually work?

Right now the situation is a mess. In the U.S., courts have ruled that AI-generated images aren't copyrightable (no human author), but haven't said whether training data is legal. The EU AI Act lays out some principles but the details aren't out yet. China's "Generative AI Services Administration Measures" require legal training data, but what counts as "legal" isn't clearly defined either.

There are a few possible paths forward:

Path one: compulsory collective licensing. Like KTV copyright — AI companies don't negotiate one by one, they just pay a collective rights management organization and get blanket licensing. Crude but effective.

Path two: expanded Fair Use interpretation. U.S. courts could rule that AI training counts as "fair use" — just like search engines indexing web pages. This path is best for AI companies but worst for creators.

Path three: revenue sharing model. AI companies share revenue proportionally with creators — the more something is used, the more it earns. Sounds great, but how do you track "how much each song contributed to the model"? That's a technical nightmare.

Whichever path we take, one thing is certain: the "wild growth" era of AI-generated content is coming to an end. The Suno incident is just the beginning. More lawsuits, more regulations, more battles are coming. AI content creators, get ready.

See you tomorrow.