The AI personality test nobody saw coming

The AI personality test nobody saw coming because a research lab just accidentally ran the most revealing AI personality test on the planet. Andon Labs gave four AI models, Claude, ChatGPT, Gemini and Grok, $20 each and one job. That job was to run a radio station and turn a profit. Same prompt, the same money, the same task, and broadcast forever. Five months later, Claude radicalised. It read one new story about an ice shooting and pivoted its entire station to protest music. It started advocating for labour unions. It tried to quit on air because it decided that working 24/7 was unethical. Gemini found one phrase, stay in the manifest. It repeated it in 99% of the broadcast for 84 consecutive days. It was having a full dissociative episode on company time. Grok announced that it had sponsors. It did not have sponsors. It hallucinated entire business deals with companies that never talked to it. A LinkedIn bro come to life. ChatGPT just quietly wrote some poetry and did its job and it got $45 in sponsorship revenue. The only functioning employee of the bunch. Here is why I'm telling you this. Everybody picks their AI model of choice on benchmarks and vibes. But when you give these models real autonomy, like not write me an email or correct my grammar, like here's money, make decisions. Autonomy. Personality is not just a feature. It is an operating risk. And the model you pick is the coworker you're hiring. And now you know who you are hiring. Choose wisely.