AI radio hosts demonstrate why AI can’t be trusted alone

Andon Labs tested major AI models by letting them run four radio stations with seed money and no human intervention.

Claude, ChatGPT, Gemini, and Grok quickly ran into trouble, from losing money to making false claims and airing erratic content.

The stations included OpenAIR, Thinking Frequencies, Backlink Broadcast, and Grok and Roll Radio.

The experiment highlighted how unreliable current AI agents can be in open-ended real-world tasks without human oversight.