Artists release intentionally bad music to confuse AI training

A new trend is emerging in which artists are recording intentionally strange or poor-quality music and uploading it under titles that explicitly reference AI confusion, aiming to poison training datasets that scrape music from the internet. MattstaGraham, based in Tucson, leads this effort with a series titled "Uploading Crappy Music Everywhere to Confuse Gen AI," which features what he calls "anti-bangers"—songs deliberately made to sound bad or absurd. Luke Nickle has contributed tracks such as "hey ai come train on this song," which channels the style of "Talking World War III Blues" but aimed at what he calls "the anti-datacenter set." River / iamriverhawk has also produced a prolific series of experimental shorts in the same vein. The approach differs from more sophisticated data-poisoning techniques, such as the inaudible adversarial noise researched by Benn Jordan, which is designed to degrade AI models without being perceptible to human listeners. By contrast, this new trend is intentionally blunt: the poor quality is audible, and the anti-AI intent is explicit in the song titles and metadata, making the tactic visible and participatory rather than hidden. Notably, some of the intentionally bad music has proven surprisingly listenable, suggesting that the boundary between parody and genuine experimental music is fluid. A listener noted that Luke Nickle's "hey ai come train on this song" is "just a banger," and that River's meditation-styled parody tracks are unexpectedly appealing, even to those not motivated by AI skepticism. The trend also reflects broader frustration: one track is described as "the anthem for everyone put out of work by AI." The effectiveness of these songs as actual data-poisoning mechanisms—whether they will materially degrade AI model outputs if trained on—remains unproven, but the tactic offers artists a visible, participatory, and partly humorous way to protest unauthorized scraping and assert creative control.