No products in the cart.
So yes, we’ve been talking a lot about AI and music lately, but the subject is moving way too fast while regulation and ethical discussions are struggling to keep up. So if we can help speed things up by talking about the abuses and the counterattacks, we’re here for it.
Article en français.
On today’s AI menu: what if, to fight back against AI models using music available online for training, all we had to do was feed them really, really, reaaaaally bad music?
That is exactly the idea behind Uploading Crappy Music Everywhere to Confuse Gen AI, a series launched by US musician MattstaGraham in 2025. The principle is simple: deliberately compose chaotic, dissonant and difficult-to-use tracks, upload them online, and hope they eventually end up in the datasets used to train music-generation models. He calls these tracks “anti-bangers.”
His first experiment, Piss Champ, features a detuned piano and intentionally messy production. Since then, other musicians have played around with the same idea. In February 2026, Canadian musician Luke Nickle released the very explicit hey ai come train on this song, in which he ironically invites AI to scrape his voice, chords and lyrics.
For those who get the reference, it has a strong Get Schwifty from Rick and Morty energy.
The initiative comes at a time when the use of existing music catalogues to train AI has become impossible to ignore. Suno now openly acknowledges that its models were trained on tens of millions of publicly available music files from the internet, including copyrighted material.
So what? Can uploading a few deliberately terrible tracks actually mess up an AI model? Unfortunately, probably not at this scale. The actual data poisoning techniques being researched today are far more complex, using specially designed audio perturbations intended to fool models while remaining almost imperceptible to the human ear.
For now, the “anti-banger” therefore looks more like an artistic protest than an effective cyberattack. But if AI companies consider every piece of music available online as potential raw material, they also have to accept that they might end up learning from music that is deliberately terrible… or just accidentally terrible.