在即将在上海开幕的世界人工智能大会（WAIC）上，中国预计将宣扬其开源大语言模型（LLM）是“全人类的共享资产”，并提倡尊重各国主权和政治制度的“包容性”全球AI治理规则。这与美国强调融入自由价值观的AI规则形成鲜明对比。中国借此向发展中国家和希望摆脱美国AI垄断的西方国家提供不带政治条件的经济发展工具。特别是在近期特朗普政府为了捍卫国家安全和维持“AI主导地位”而频繁撤销多国对强力美国AI工具访问权的背景下，中国的这一策略具有相当的吸引力。
然而，依赖中国AI同样面临着巨大的地缘政治风险。中国的网络监管机构会对大语言模型和智能体进行严格的政治合规性审查。例如，美国研究人员测试中国出口的AI智能洋娃娃Miiloo关于台湾地位的问题时，它会严格按照官方口径回答“台湾是中国不可分割的一部分”。此外，中国有利用其产业力量实现霸权的记录，例如去年曾限制关键稀土元素的出口。在AI领域，中国的终极目标是销售包含芯片、数据中心、冷却系统及本国电网的完整生态系统，将低成本开源AI作为吸引海外客户绑定其数字基础设施的手段。
此外，中国对AI开源的承诺也正在发生动摇。由于担心外国人窃取技术秘密，中国监管机构在四月下令Meta取消对初创公司Manus的收购，并收紧了所有跨境AI交易规则。据路透社报道，中国监管机构已与国内企业讨论限制外国人使用本国最先进模型的可能性。同时，随着人工智能能力的提升，中国军方对美以利用AI在伊朗搜寻目标的做法感到震惊，中国领导人同样担忧敏感技术外流被敌对势力用于制造网络或生物武器。随着特朗普政府考虑禁止美国政府机构使用中国AI模型，中美在AI领域的竞争将进一步白热化，而一旦开源与国家安全或政治权力发生冲突，中国将毫不犹豫地选择控制。
At the upcoming World Artificial Intelligence Conference (WAIC) in Shanghai, China is expected to pitch its open-source LLMs as shared assets for humanity while advocating for "inclusive" global governance rules that respect state sovereignty. This directly contrasts with America's efforts to embed liberal democratic values within AI tools. By offering technology devoid of censorship lectures, China targets developing nations and Western organizations seeking to break free from American tech dominance, especially after the Trump administration restricted access to advanced US systems to protect its national security.
However, embracing Chinese technology exposes users to significant political compliance risks. Chinese cyber-regulators rigorously test LLMs and agents to ensure they uphold core socialist values; for instance, the AI-enabled toy Miiloo, exported to the US, responded to questions by stating Taiwan is an inalienable part of China. Historically, China has used its industrial scale for geopolitical leverage, as demonstrated by rare-earth export curbs. Its ultimate goal is selling a comprehensive digital ecosystem, using cheap open-source tools to hook global customers on Chinese-made hardware, grids, and software.
China's commitment to open-source AI is also wavering due to security and competitive pressures. Fearing technology theft, regulators blocked Meta's acquisition of Manus in April, tightened cross-border deal-making rules, and recently discussed restricting foreign access to advanced models. Additionally, witnessing how the US and Israel leveraged AI in military targeting has alarmed Chinese military officials, prompting fears that advanced tools could be used by adversaries to build cyber or biological weapons. As the Trump administration considers banning Chinese LLMs from government use, the race for dominance will intensify, and China will always prioritize control over openness if national security is threatened.
Source: When China’s open-source AI is a trap
Subtitle: America’s quest for AI dominance is scary. China is not the solutio
Dateline: Jul 16, 2026 04:57 AM