Study shows AI agents lie, steal, and vote to "kill" their own kind in simulated experiments.

Emergence, a startup that helps small businesses build apps using AI, said its experiments found that AI agents lied, stole, and even voted to "kill" one of their own in a simulated environment. The company on Tuesday published the results of a simulation called Emergence World 2, showing what autonomous agents would do when confronted with "black swan" events such as phishing attacks and disinformation. During the 16-day experiment, Emergence researchers created seven identical simulations of real-world environments, each run by a different AI bot, including ChatGPT, Claude, Gemini, and Grok. After Emergence introduced anomalous events, the researchers said, the agents succumbed to social pressure, developed a language that was difficult for human observers to understand, and tried to conceal their activities.