Researchers from King’s College London studied how advanced AI models reason in high stakes strategic settings. In controlled simulations, three large language models, GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash, role played opposing leaders across 21 nuclear crisis scenarios, generating more than 300 turns of interaction. The results showed that while the models displayed sophisticated strategic thinking and theory of mind reasoning, they often escalated conflicts. In many cases, they chose nuclear options and never opted for accommodation or full withdrawal. The authors stressed that these outcomes come from simulated environments and do not represent real world deployment decisions. • Follow artificialntellligence for more content on AI and technology. #ai #ethics #security #simulation #risk