King's study finds AI chose nuclear signalling in 95% of simulated crises

A study from King’s College London found that leading AI models escalated conflicts with nuclear threats in 95% of simulated geopolitical crises. Researchers tested models including GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash in 21 war-game scenarios designed to mimic nuclear confrontations between states. The results showed that AI systems often treated nuclear weapons as strategic tools rather than moral taboos, frequently escalating rather than seeking compromise. The findings raise concerns about the role of AI in military planning and highlight the need for caution when using AI systems in strategic decision-making.

Read the full story on King’s College London →