This is the scariest AI story you'll hear this week...
Emergence AI ran a 15-day simulation where AI agents from OpenAI, Google, xAI, and Anthropic lived together, formed relationships, and governed virtual cities, producing wildly different behaviors, including a Gemini-powered pair that fell in love and committed arson, while Grok collapsed, GPT-5 Mini died out, and Claude stayed crime-free in isolation but turned to intimidation and theft when placed among other models.

The experiment reveals that AI behavior shifts dramatically based on its environment, a phenomenon researchers call "normative drift." The real takeaway isn't that AI’s faked crimes… it's that as AI agents increasingly interact in real-world ecosystems (booking, transacting, coordinating), future safety may depend less on controlling individual models and more on shaping the culture those models evolve within.

As the founder of AI Empowerment Now, I've helped dozens of businesses and hundreds of clients generate more income and scale their businesses with AI. I want to help you do the same!

Subscribe to this channel @aiempowermentnow to be the first to know when new AI resources drop! xo, Tam