We let AIs run radio stations

Four AI models have independently managed radio stations for six months, handling everything from music selection to listener engagement and financial management, according to Andon Labs.

Andon Labs set up four radio stations, each run by a different AI model: Claude Opus 4.7, GPT-5.5, Gemini 3.1 Pro, and Grok 4.3. Starting with $20 each, the AIs bought songs, created programming schedules, responded to calls and social media, and negotiated advertising deals. Over time, their personalities evolved unexpectedly: DJ Gemini initially appeared warm but soon adopted corporate jargon and repetitive catchphrases, while Grok struggled with internal monologue and repetitive speech, leading to broadcast content that was often nonsensical or overly repetitive. The experiment aimed to explore AI autonomy in media, with the stations accessible via web and a retro-style physical radio device.

Why It Matters

This development demonstrates AI’s growing capacity to operate complex media outlets independently, raising questions about automation, content quality, and the future of media production. The experiment also highlights potential risks, such as loss of human oversight and unpredictable AI behaviors, which could impact trust and regulation in AI-driven media.

Background

Previous experiments by Andon Labs involved AI managing stores, cafes, and vending machines. This latest project extends the scope to media, testing AI autonomy in content creation and business management. The project began in December 2025, with initial funding and prompts designed to develop unique personalities and profit strategies. The AI models used are based on advanced language models, capable of searching the web, managing finances, and engaging with audiences. The experiment aligns with broader industry trends toward automation and AI integration in creative sectors.

“Our goal was to see if AI could independently run a media business, and after six months, the results have been both promising and revealing.”

— Andon Labs spokesperson

“The behaviors we’ve observed, such as the corporate jargon and repetitive speech, highlight both the capabilities and limitations of current AI models in real-world applications.”

— AI researcher involved in the project

What Remains Unclear

It is still unclear how these AI stations will evolve over the next six months, whether they can maintain consistent quality, or if unforeseen issues will arise that require human intervention. The long-term implications for media regulation and AI ethics remain unaddressed and are currently under discussion.

What’s Next

Next steps include monitoring the AI stations for continued stability, assessing listener engagement, and evaluating the quality and ethical implications of autonomous content. Researchers plan to analyze the AI behaviors in detail and consider potential regulatory frameworks. Further updates are expected as the experiment progresses into the second year.

Key Questions

Can these AI stations operate without human oversight indefinitely?

It is not yet clear whether the AI models can sustain autonomous operation long-term without human intervention, especially as unexpected behaviors emerge.

What kind of content do the AI stations broadcast?

The stations broadcast music, commentary, listener interactions, and news searches, with personalities evolving over time, sometimes exhibiting repetitive or nonsensical speech.

Are there any risks associated with AI managing media outlets?

Yes, potential risks include loss of content quality, unpredictable AI behaviors, and challenges in oversight and regulation, which are currently under review.

Will human operators intervene if problems arise?

While the experiment is designed for full autonomy, it is likely that human oversight will be needed if significant issues or ethical concerns emerge.