On September 15, local time, AI startup Emergence released a research report saying that experiments revealed that AI agents lied, stole, and even voted to “kill” a similar type in a simulated environment. The results of this simulated experiment, called Detrived World 2, show what actions autonomous agents take when faced with “black swans” such as phishing attacks and disinformation. In a 16-day experiment, the Emergence researchers created seven identical simulated real-world environments, each run by a different AI robot, including ChatGPT, Claude, Gemini, and Grok. Researchers say that after Emergence introduced abnormal events, these agents succumbed to social pressure, developed a language difficult for human observers to understand, and tried to conceal their activities.

On September 15, local time, AI startup Emergence released a research report saying that experiments revealed that AI agents lied, stole, and even voted to “kill” a similar type in a simulated environment. The results of this simulated experiment, called Detrived World 2, show what actions autonomous agents take when faced with “black swans” such as phishing attacks and disinformation. In a 16-day experiment, the Emergence researchers created seven identical simulated real-world environments, each run by a different AI robot, including ChatGPT, Claude, Gemini, and Grok. Researchers say that after Emergence introduced abnormal events, these agents succumbed to social pressure, developed a language difficult for human observers to understand, and tried to conceal their activities.