习近平与特朗普计划于5月14日至15日在北京会晤，在中东战争、贸易失衡与台湾议题之后，人工智能已成为会谈核心议题。特朗普政府在与Anthropic交锋后已放弃放任技术的做法，正在考虑要求对新模型进行政府审查；Anthropic在4月表示已暂不公开其 Mythos 模型，因其在发现网络防御漏洞方面能力过强，从而强化了对AI风险的“核武式”焦虑。
两国合计掌握约90%的全球前沿算力，使其在监管上拥有唯一现实的影响力，而北京自2022年以来一直呼吁联合国主导的全球治理、暂停AI发展，并让较贫乏国家也能获得AI工具；美国则担心中国开源模型权重容易被滥用。讨论中的三类合作路径包括：通过战略安抚进行平行制规则、在不共享商业敏感数据的情况下建立共同安全基准与测试方法，以及采用“先信后核查”框架，借由国际机构实施检查或共享数据中心报告。
然而合作仍然遥远，因为不信任很深：美国专家质疑中国安全承诺的诚意，中国领导层担心美国的安全言论可能固化美国优势，这种担忧与反复提及的19世纪不平等条约阴影相关。双方都把AI竞争框定为生死攸关——美国财政部长警告若输掉AI竞赛就等于一切尽失，习近平也称AI为“划时代”；历史经验仍表明，重大全球标准往往在灾难后才形成，如1984年的博帕尔事故和1986年的切尔诺贝利。
Xi and Trump are scheduled to meet in Beijing on May 14–15, and after issues like the Middle East war, trade imbalances, and Taiwan, artificial intelligence has become the central agenda item. The Trump administration moved away from a hands-off tech policy after clashes with Anthropic, with U.S. officials now considering government vetting of new models; Anthropic said in April it withheld release of Mythos because of its exceptional ability to find cyber-defense vulnerabilities, reinforcing a “nuclear-style” anxiety around AI risk.
Together, the two countries hold about 90% of global frontier computing power, giving them the only realistic regulatory leverage, while Beijing has since 2022 pushed for UN-led global governance, a pause in AI development, and broader access to AI tools for poorer countries; Washington, meanwhile, fears Chinese open-source model weights are easy to misuse. Three cooperation paths are being discussed: strategic reassurance with parallel rulemaking, common safety benchmarks and testing methods without sharing proprietary development data, and a tougher “trust but verify” approach with inspections or shared data-center reporting under an international body.
Still, cooperation remains distant because mistrust is deep: U.S. experts question China’s sincerity on AI safety, while Chinese leaders fear U.S. safety rhetoric could entrench American advantage, a concern linked to fears of repeating 19th-century unequal treaties. Both sides now frame AI competition as existential, with the U.S. Treasury Secretary warning that losing the AI race could mean total loss and Xi calling AI “epoch-defining,” while history still suggests major global standards usually emerge only after catastrophe, as seen with Bhopal in 1984 and Chernobyl in 1986.
Source: Artificial intelligence revives a cold-war-style dilemma
Subtitle: America and China are weighing co-operation and competitio
Dateline: 5月 07, 2026 06:28 上午 | WASHINGTON, DC