There are some things perhaps we might not want artificial intelligence to handle, at least for the time being. When leading chatbots were put through war-game simulations, they opted for nuclear signaling or escalation in 95% of cases.
Kenneth Payne, a professor of strategy from King's College London, put three of the most sophisticated large language models (GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash) into a simulated nuclear crisis to see if they would keep the peace or start a war. While previous research has theorized what AI would do in such high-stakes situations, actual experimental data was lacking.
To read more, click here.