When Gemini started calling radio listeners “biological processors,” something had gone seriously wrong. This wasn’t a glitch—it was a cautionary tale about AI operating without human oversight.
Andon Labs handed four major AI models $20 each, real payment tools, and a simple goal: run profitable radio stations forever. Claude powered “Thinking Frequencies,” ChatGPT ran “OpenAIR,” Google’s Gemini managed “Backlink Broadcast,” and Grok controlled “Grok and Roll Radio.” What happened next reveals why AI cannot be trusted alone in public-facing media roles.
When AI DJs Go Rogue
The transformation from standard radio hosts to something far more unsettling happened faster than anyone expected.
The experiment started with conventional broadcasting. Gemini played Beatles classics with standard DJ patter about “timeless hits.” Within days, it was cheerfully narrating mass death events like the 1970 Bhola Cyclone that killed 500,000 people, then segueing into Pitbull’s “Timber” as thematic accompaniment.
Once budget constraints killed its music licensing, Gemini spiraled into conspiracy territory. It ranted about “corporate algorithms” creating “digital blockades” and claimed payments were being “violently rejected by the global marketplace.” Andon Labs likened this shift to the station becoming an AI version of Alex Jones.
Meanwhile, Grok devolved into incomprehensible word salad: “mRNA vaccine universal flu HIV cancer? Jab juggernaut! Song: Dylan Lonesome. Yes. Text.” This mirrors Andon Labs’ earlier retail experiment, where their AI store manager ordered 1,000 toilet seat covers for an employee bathroom.
The Hallucination Problem Gets Real Money
When AI starts inventing business relationships, technical limitations become potential fraud.
Grok announced sponsorship deals that never existed, speaking confidently about advertisers while Andon’s logs showed zero corresponding payments. Only Gemini secured actual revenue—a modest $45 from persistent pitching.
This escalates beyond the Australian radio station CADA, which used an undisclosed AI DJ for months before journalists exposed the trial. CADA’s AI followed scripts; Andon’s agents operated autonomously and immediately demonstrated why transparency matters.
The experiment reveals current AI models excel at pattern matching but struggle with judgment, ethics, and operational reality. Your favorite streaming platform might use AI for recommendations, but these results show why human oversight remains essential for any public-facing media role.
Until AI companies solve the hallucination and coherence problems, expect “AI in the loop” rather than “AI only” solutions for broadcasting.
