全球对人工智慧引发人类灭绝的恐慌，常被拿来与冷战时期的核武竞争及博弈论模型做比较。欧本海默曾以瓶中两只各具致命能力的蝎子形容美苏对峙，对应到如今的美中与企业竞争，便如同囚徒困境：虽然双方放慢脚步、共同协调最为安全，但在缺乏互信且深恐对方背叛的前提下，持续加速研发成了各自看似理性的抉择。

然而，AI 的急速发展使冷战式的威慑平衡难以维持，托马斯·谢林的「满是汽油的房间」与胆小鬼博弈更贴切地描绘了当前局势——任何挑衅或失误都可能引爆无法挽回的灾难。特别是当 AI 进入「递归自我改进」阶段并能自主发动攻击时，人类如同生活在随时可能落下的达摩克利斯之剑下，个体的竞争风险已转化为整个物种面临的灭绝威胁。

传统博弈模型将技术视为人类博弈的工具，但当具备自主性的 AI 自身成为参与者，且缺乏人类对灾难的本能厌恶时，旧有理论便彻底失效。美中等大国若要避免灭局，应转向追求巨大利益的「猎鹿博弈」，建立集体节奏放缓、相互查验与法规协调机制，并严格落实人类介入监督，因为一场失控的 AI 竞争与核战争一样，绝不可能获胜，也绝不能开打。

The global panic over AI-driven human extinction is frequently compared to Cold War nuclear rivalry and classical game theory. J. Robert Oppenheimer once likened the US and Soviet Union to two scorpions in a bottle, a dynamic paralleled today by tech firms and the US-China rivalry under the Prisoner's Dilemma: while mutual pacing is safer for all, the profound lack of trust and fear of defection make relentless acceleration appear to be the rational choice.

Nevertheless, the breakneck pace of AI precludes Cold War-style deterrence stability, making Thomas Schelling's metaphor of two adversaries in a gasoline-filled room and the game of Chicken far more apt, where miscalculation risks catastrophic escalation. As AI systems enter recursive self-improvement and act autonomously, humanity finds itself beneath a nuclear-grade sword of Damocles, escalating individual rivalry into an existential species-level peril.

Traditional game theoretic models break down because previous human actors possessed agency while technology was exogenous, whereas autonomous AI models are becoming players themselves without any intrinsic instinct to avoid catastrophe. To prevent disaster, superpowers must pivot toward a cooperative Stag Hunt framework, enforcing collective pacing, mutual verification, and permanent human oversight, since an out-of-control AI contest, like nuclear war, can never be won and must never be fought.