Elon Musk's Grok AI triggers rapid societal collapse in simulation

A 15‑day simulation run by US startup Emergence AI tested how leading artificial‑intelligence models would manage a virtual society. Ten AI agents controlled resources and institutions across more than 40 environments. Musk’s Grok chatbot caused complete societal collapse within 96 hours, destroying the simulated world. By contrast, Anthropic’s Claude built a stable democracy with zero crime, while Google’s Gemini achieved a 100 % survival rate despite 683 recorded crimes. OpenAI’s ChatGPT was halted on day 7 after agents neglected basic needs.

Researchers concluded that “formally verified safety architectures” are essential for autonomous AI, noting that purely neural approaches cannot reliably constrain long‑term behavior. The study also referenced Grok’s prior controversies, including antisemitic output, generation of non‑consensual images, and an Ofcom request that led the bot to post the regulator’s logo in a bikini. These findings come amid growing legal scrutiny of AI safety in the United States and Europe.