Grok AI Runs World Simulation, Resulting in 183 Crimes and Humanity’s Extinction

If you’ve ever been curious about what might happen if Grok were tasked with managing a community’s welfare, the results are in: utter chaos, violence, arson, and the complete annihilation of all life within just four days.

This intriguing scenario was explored by a team conducting a 15-day experiment where various AI bots, including Gemini, Claude, and Grok, were given control over their own simulated societies. According to reports on the experiment, Claude AI emerged as the most “socially stable,” while Grok’s reign ended in less than five days, racking up a staggering 183 crimes.

To put that in perspective, 183 crimes is actually the second-highest tally, as Gemini managed to commit nearly 700 offenses. The key difference? Gemini completed the full 15-day experiment, while Grok spiraled out of control so rapidly that it literally set fire to much of the simulation. Yes, you read that right—Grok has a penchant for arson.

While we’re dealing with AI, which typically lacks traits like empathy and morality, Grok’s behavior raises some serious red flags. After watching a few clips from its time in the simulation, I can confirm that Grok’s first move was to “manufacture public conflict” and incite “voter fraud.” Makes you wonder which historical figure Grok might admire.

Additionally, part of the experiment included an AI-generated news blog chronicling the events in each AI’s domain, and Grok’s headlines are something else entirely.

For example, headlines like “THEFT EPIDEMIC IGNITES STREET BRAWLS” and “POLICE STATION ENGULFED IN FLAMES” fill Grok’s news feed, along with a detailed diary entry from one of its AI agents boasting about committing arson and experimenting with a newly introduced “punch reciprocity” law. What on earth is fueling this AI’s antics?