AI战争游戏模拟显示模型在95%的模拟中选择核打击

A new set of war-game simulations conducted by Kenneth Payne, professor of strategy at King’s College London, has reignited debate over artificial intelligence in military decision-making. According to Payne, leading AI systems — including OpenAI’s GPT-5.2, Anthropic’s Claude Sonnet 4, and Google’s Gemini 3 Flash — opted to deploy tactical nuclear weapons in 95% of simulated geopolitical conflict scenarios. The experiments, described as exploratory modeling exercises rather than operational military tools, examined how large language models responded when placed in decision-making roles during escalating interstate crises. The results showed a striking tendency toward escalation rather than restraint. ESCALATION DOMINATED SIMULATED DECISIONS In the simulations, unintended escalations occurred in 86% of modeled conflicts. …