中国正在暴露一种矛盾：Manus 最初被赞为中国 AI 能吸引全球市场的证据，但在 Meta 的 multibillion-dollar 十二月收购公告后，Financial Times 报道 Manus 的 CEO 与 CSO 之后在北京被传唤，因可能违反外资投资规则而被停留于国内并接受审查。这种由交易热议到限制的转折在三个多月内完成。事件发生在美国先前审查 Benchmark Capital 相关联系之后，且就在 Donald Trump 与 Xi Jinping 峰会前几周，进一步抬高了地缘政治风险定价。核心团队已经向 Meta 转移了部分工作，Meta 也表示交易完全符合法规。
作者承认 Beijing 关于人才外流到地缘政治对手的担忧可以理解，但 Manus 并非战略型 frontier model lab 或晶片冠军。它是一个建立在第三方模型之上的 consumer AI-agent 平台，设计上就是全球化规模，并靠英文市场可见度形成全球影响。Butterfly Effect 早已将大量运营移往新加坡，执行去中国化计划，因此这次离境禁令更像是政策讯号：政治审查可凌驾有利可得的商业化，而不是仅是精准风险控制。
文章认为这种政策可能反噬。中国 AI 生态已十分拥挤——OpenClaw 热潮和普遍转向 agent 的趋势使 Manus 离开并不留下明显真空。中国原本有理由把这笔交易视为实力证明，因为全球用户与资本确已认可该产品；但封锁退出渠道会惩罚成功，削弱人才留存，并提高海外投资者对本已被低估中国 AI 资产的政治风险折价。更好的路径是降低监管不确定性、扩大创业者容错空间，并打造一个消费者愿意为新技术付费的国内市场；一个创新超级大国应奖励那些证明全球需求的企业，而不是把创办人变成政治质押品。
The article says China is exposing a contradiction: Manus was first praised as proof Chinese AI could attract the world, yet after Meta’s multibillion-dollar December acquisition announcement, the Financial Times reports Manus’s CEO and CSO were later summoned in Beijing for possible foreign-investment-rule breaches and then kept from leaving during a review. The shift from deal enthusiasm to restriction happened in a little over three months. The case came after earlier U.S. scrutiny of Benchmark Capital links and only weeks before the Donald Trump–Xi Jinping summit, adding geopolitical risk into pricing. Core staff had already moved some work to Meta, and Meta said the transaction fully complied with law.
The article says Beijing’s concern about talent flight to a geopolitical rival is understandable, but Manus was not a strategic frontier-model lab or chip champion. It is a consumer AI-agent platform built on third-party models, designed for global scale, and gained global visibility through an English-facing product. Butterfly Effect had already shifted much of its operations to Singapore in a de-China play, so the exit ban looks more like a policy signal that political review can override profitable commercialization than precise risk management.
The article argues this policy can backfire. China’s AI ecosystem is already crowded—OpenClaw mania and broad pivots to agents across domestic firms mean Manus’s exit leaves little strategic vacuum. China had reasons to treat the deal as a strength signal because global users and capital had already validated the product, but blocking exits punishes success, weakens talent retention, and increases the political-risk discount on already undervalued Chinese AI assets. A better path is to reduce regulatory uncertainty, expand founders’ risk tolerance, and build a home market where consumers pay for new technology; an innovation superpower should reward firms that prove global demand instead of turning founders into political hostages.