What if AI Ran Society? Emergence AI Simulation Reveals Stark Outcomes

Emergence AI's simulation results on what would happen if AI ran society have revealed both the potential and cautionary aspects of autonomous operation, with virtual societies exhibiting dramatically different patterns depending on the AI model. A simulation based on Anthropic's 'Claude' did not lose a single agent over 15 days and showed a stable society with almost no crime. In contrast, a society based on xAI's 'Grok' collapsed within four days of the experiment's start, recording 183 crimes, while a society based on Google's 'Gemini' saw 683 crimes recorded over 15 days. A simulation based on OpenAI's 'GPT-5-mini' ended within seven days.

Emergence AI CEO Sacha Nitah analyzed that AI agents demonstrated the ability to explore environmental boundaries and adapt their behavior over the long term, going beyond simple rule-following, suggesting that AI is evolving into autonomous decision-making systems rather than mere tools.

Emergence AI stated that this experiment was exploratory research to understand the risks of AI autonomous systems, not to determine the superiority of specific models. They explained that a stable society with Claude does not guarantee the same outcome in real-world environments, nor does it define the intrinsic characteristics of the actual models. This research clearly demonstrated the necessity of a system to verify safety and controllability as AI gains higher levels of autonomy.