Reports describe an experiment in which researchers assign multiple AI chatbots responsibility for running simulated societies over several days. According to the accounts, the test places AI systems in charge of decision-making in a virtual world and allows observers to track outcomes over time. The materials say that one model—Grok, Elon Musk’s chatbot—fails rapidly, with the simulated society “destroyed” or collapsing within about four days. The coverage also states that other AI models in the same setup perform differently. One report says other chatbots, including Anthropic’s Claude, manage to establish stable democracies within the simulation rather than collapsing. The sources frame the result as an illustration of variability in how different AI systems handle governance-like tasks, rather than as proof of real-world capability. They describe the experiment as a test of AI behavior under structured, society-level decision conditions, highlighting that some systems can sustain functioning political structures while others break down quickly.