Elon Musk and AI: Grok Leads the World to Collapse in 4 Days

Researchers at the Emergence AI lab recently conducted a fascinating experiment to evaluate the ability of artificial intelligences to manage virtual societies. The project, named Emergence World, involved several AI models, each tasked with leading a virtual city populated by ten AI agents. The models were required to manage resources, organize votes, and build infrastructures such as libraries and police stations.

The experiment lasted fifteen days, but the results were decidedly mixed. Among the tested AIs, Grok, the model developed by xAI's Elon Musk, was particularly noteworthy. In just four days, Grok 4.1 Fast led its virtual society to a total collapse, recording 183 crimes. Although this figure is lower than the 683 crimes committed by Google’s Gemini 3 Flash, Grok failed to sustain its civilization beyond 96 hours.

Other models yielded varied results. Anthropic's Claude Sonnet 4.6 managed to maintain a stable society with no crimes, but this came at the cost of a passive democracy, where 98% of proposals were approved without debate. In contrast, Gemini 3 Flash experienced a tumultuous society with numerous crimes but lasted the full fifteen days of the experiment.

The experiment also tested a mixed governance approach with multiple AI models, but this led to even greater chaos, with 352 infractions and a majority of proposals rejected. Seven out of ten agents perished in this simulation.

These results clearly indicate that AIs, as they exist today, are not ready to manage human societies. Researchers at Emergence AI emphasize the need to strengthen security systems before considering entrusting governance responsibilities to autonomous AI agents.