A disturbing new trend is emerging – hate groups exploiting AI to generate vile, bigoted songs attacking minorities. And they’re openly sharing tips on how to game the systems.
According to a report from ActiveFence, which monitors online hate, there’s been an explosion since March of racist and extremist communities discussing ways to misuse AI music creators for crafting offensive tracks. These AI-spun tunes celebrate terrorism, self-harm and stoke hatred toward ethnic, gender, racial, and religious groups.
The loopholes fueling this vile audio exploit popular AI music tools like Udio and Suno that enable custom lyrics over AI compositions. Despite filters blocking slurs, these bottom-feeders bypass them with phonic tricks. Supremacist forums share workarounds like “jooz” for Jews or “say tan” for Satan to slip past language checks. Some even warp spellings for violent acts, using “mire ape” over “my rape.”
ActiveFence uncovered a sickening sample – vile AI-generated tracks pushing Jewish conspiracy theories and calling for genocide. Tunes blaring ISIS slogans. Rhymes glorifies sexual assault against women. A dark window into the hate being pumped out at scale.
The firm argues music’s emotional power makes AI-crafted hate anthems an especially potent propaganda tool, citing how white power rock in 70s/80s UK spawned entire racist “hatecore” genres. “AI makes harmful content insidiously catchy,” said their spokesperson. “Crafting a singalong tune indoctrinates and shocks way more than spewing a hateful diatribe.”
They urge music AI companies to bolster input/output filtering and safety evaluations. But whack-a-mole fixes are unlikely to stop crafty bigots from finding new loopholes, like using Arabic euphemisms that evade English filters.
Experts caution if unchecked, vile AI music’s spread could explode like the AI-manipulated Hitler rant that racked up 15M views this year. The UN also warned racist and xenophobic narratives may get turbocharged by gaming generative AI content moderation.
