矽谷科技领袖近期转变论调，呼吁美中两国合作以减缓人工智能竞赛步伐。Anthropic 执行长达里奥·阿莫代伊提出「调整前沿步伐」策略，主张全球协调需要与中国合作，但同时强调美国及民主阵营必须维持最大领先优势，并采取措施如打击晶片走私与限制未授权模型蒸馏来削弱中国 AI 产业。然而，中国政府与科技界对此提议深感怀疑，认为这是一项旨在先行限制中国并极大化美方筹码的不平等交易，缺乏对等谈判的诚意。

中国方面对美方呼吁的保留态度源于双方对风险定义与动机的不同认知。与矽谷普遍担忧的灭绝性生存风险相比，北京政府更聚焦于 AI 对国内社会稳定、国家安全及政权统治带来的具体威胁，如政治虚假讯息、网路攻击与社会动荡。此外，中国决策者与产业界普遍认为美方实验室借安全之名行打压对手或拉抬估值之实，而中国业界（如 DeepSeek 等研究者）则倾向透过开源与低成本模式推广技术普及，反对由少数美国科技巨头垄断前沿 AI 资源。

尽管彼此存在互信障碍与言论争议，美中两国在 AI 安全领域的合作仍未完全破裂。随著即将展开的高层会晤，双方初步协商已逐步启动，且两国在网路安全、强大模型滥用及系统失控等议题上具备共同关切点。专家指出，只要建立可靠的双边履约验证机制，聚焦于具体范畴展开善意对话，仍有望取得实质进展；但美方领袖若持续发表削弱对手的言论，恐将进一步损害外交谈判所依赖的脆弱信任基础。

Silicon Valley tech leaders have recently shifted their messaging to urge U.S.-China cooperation in slowing down the AI race. Anthropic CEO Dario Amodei proposed a 'pacing the frontier' strategy, arguing that global coordination requires China's involvement, yet simultaneously insisting that the U.S. and democratic allies must maintain a significant technological lead by curbing chip smuggling and preventing model distillation. However, Beijing and Chinese tech firms view this proposal with deep skepticism, perceiving it as a lopsided deal aimed at subordinating China and maximizing U.S. leverage before any talks begin.

China's hesitation stems from fundamentally divergent views on risks and motivations compared to U.S. labs. Rather than prioritizing catastrophic existential threats of human extinction, Chinese authorities are far more concerned with tangible disruptions to domestic stability, cybersecurity, and political governance, such as disinformation and espionage. Furthermore, Chinese policymakers and industry leaders suspect that American safety rhetoric serves instrumental purposes—such as inflating corporate valuations or suppressing competitors—while researchers in China advocate for open-weight, accessible AI to prevent a handful of Western giants from monopolizing frontier technology.

Despite significant friction and rhetoric, avenues for U.S.-China AI safety collaboration remain open. With high-level bilateral summits on the horizon, initial discussions are already underway around shared concerns, including cyber threats, the misuse of powerful systems, and unforeseen model behaviors. Analysts emphasize that while progress is possible through narrow, good-faith dialogue and verifiable compliance frameworks, confrontational rhetoric advocating the containment of China's AI ecosystem risks undermining the fragile diplomatic trust necessary for meaningful global governance.