分析人士表示，中国提出的人工智能技术出口禁令和管控措施旨在加强北京在该战略性关键领域的控制，给国内模型开发商在全球人工智能竞赛中的商业前景蒙上阴影。
China’s proposed export bans and controls of artificial intelligence technologies aim to tighten Beijing’s grip in the strategically vital sector, casting a cloud over the commercial prospects of domestic model developers in the global AI race, analysts say.
这些新措施的出台正值全球对先进人工智能系统风险的担忧日益加剧之际，此前发生了一系列近期网络攻击事件，同时也凸显了人工智能对北京和华盛顿日益增长的国家安全重要性。
The new measures come as concerns mount globally over the risks of advanced AI systems, following a string of recent cyberattacks, while also underscoring AI’s growing national security importance for both Beijing and Washington.
咨询公司iiMedia的创始人兼首席分析师张毅表示，新措施将不可避免地为中国企业带来短期适应成本，而此时他们正瞄准国际市场。
Zhang Yi, founder and chief analyst at consultancy iiMedia, said the new measures would inevitably introduce short-term adaptation costs for Chinese companies, at a time when they were targeting international markets.
根据中国商务部（MOFCOM）和科技部周五联合发布的草案方案，将对23项人工智能相关技术实施出口禁令和限制，包括模型、智能体和芯片设计。还提出了对生物技术和光伏行业关键技术的出口禁令和限制。
Under the draft plan, jointly issued by China’s Ministry of Commerce (MOFCOM) and the Ministry of Science and Technology on Friday, export prohibitions and restrictions would be imposed on 23 AI-related technologies, including models, agents and chip designs. Export bans and restrictions on critical technologies relating to the biotechnology and photovoltaic sectors were also proposed.
对于基础人工智能模型——该领域进步的核心驱动力——当局列出了若干触发限制的阈值。这些包括：在中国重要数据上训练的模型；算力达到10²⁶ FLOPs的模型——这是一个广泛使用的阈值，用于表示前沿人工智能模型；参数规模达到2万亿或以上的模型；或在全球前十名排行榜中的模型。
For foundational AI models, the core drivers of advancements in the sector, the authorities outlined several thresholds for restrictions to be imposed. These include models “trained on China’s important data”, developed with computing power reaching 10²⁶ FLOPs – a widely used threshold denoting frontier AI models – sized at 2 trillion parameters or more, or ranking in a global top 10 leaderboard.
随着两国试图在激烈的商业和地缘政治竞争中寻找共同应对日益增长的人工智能风险的途径，人工智能日益成为中美关系的核心。
AI has grown increasingly central to China-US relations as the two countries seek to work out ways to jointly address growing AI risks in the midst of fierce commercial and geopolitical competition.
中国国家主席习近平和美国总统唐纳德·特朗普在上个月华盛顿的峰会上同意建立双边对话，以管理人工智能风险。人工智能预计将成为他们今年剩余两次会晤的主要议程项目——下个月在深圳举行的亚太经合组织（APEC）峰会以及12月在迈阿密举行的二十国集团（G20）峰会。
Chinese President Xi Jinping and US President Donald Trump agreed to set up a bilateral dialogue to manage AI risks during last month’s summit in Washington, with AI expected to be a major agenda item at the two remaining meetings between the pair this year – at the Asia-Pacific Economic Cooperation (Apec) summit in Shenzhen next month and the G20 summit in Miami in December.
然而，张表示，北京对该行业日益加强的审查不应被解读为试图阻止中国公司在全球部署其人工智能服务的努力。
Still, Beijing’s growing scrutiny of the industry should not be interpreted as an effort to prevent Chinese companies from deploying their AI services globally, according to Zhang.
“关键在于保持核心工程技术受控，同时保持商业服务开放，”他说。
“The key lies in keeping core engineering technologies controlled, while keeping commercial services open,” he said.
商务部发言人也持相同评价，该发言人周五表示：“此次监管所涵盖的人工智能技术范围是有针对性且有限的。”
The assessment was echoed by a MOFCOM spokesperson, who said on Friday that “the scope of AI technologies covered by the regulation this time is targeted and limited”.
发言人试图澄清，中国科技公司提供人工智能服务海外的一些方式，包括通过官方应用程序编程接口和移动应用，将不受新出口限制的约束。
The spokesperson sought to clarify that some ways in which Chinese technology companies are providing AI services overseas, including through official application programming interfaces and mobile apps, would not be covered by the new export restrictions.
“这些措施正式生效后，绝大多数人工智能技术仍将可以自由出口，”发言人说。
“Once the measures take formal effect, the vast majority of AI technologies will still be freely exportable,” the spokesperson said.
一个关键的不确定性在于，开放权重人工智能技术——中国在此领域全球领先——在最终确定的出口制度下将如何被处理。
A key uncertainty is how open-weight AI technologies, an area in which China leads globally, would be treated under the finalised export regime.
越来越多的中国开放权重人工智能开发者，如开发Kimi模型的Moonshot AI和开发GLM模型的Z.ai（亦称智谱AI），正与亚马逊网络服务等主要美国超大规模云服务商签订收益分成协议，以在其云基础设施上部署其开放权重模型，这代表了一项新的重要收入来源。
Growing numbers of Chinese open-weight AI developers such as Kimi model maker Moonshot AI and GLM maker Z.ai, also known as Zhipu AI, are entering into revenue-sharing agreements with major US hyperscalers such as Amazon Web Services to deploy their open-weight models on their cloud infrastructure, representing a new major source of revenue.
根据新措施目前的措辞，若干中国人工智能模型——这些模型是全球领先的开放权重系统之一——将符合技术门槛。
Under the current wording of the new measures, several Chinese AI models, which are some of the world’s leading open-weight systems, would meet the technical thresholds.
然而，商务部发言人周五表示，中国仍在鼓励开源人工智能，新的许可要求将被加快为“即时备案即放行”。
However, the MOFCOM spokesperson said on Friday that China was continuing to encourage open-source AI, with new licensing requirements to be expedited as “immediate clearance upon filing”.
发言人称，将各种人工智能技术列入出口管制清单旨在“确保人工智能技术的安全性和可控性”，以应对日益增长的风险，例如“非国家行为者和恶意实体”滥用强大人工智能系统。
The addition of various AI technologies to the export control list was intended to “ensure the security and controllability” of AI technologies amid growing risks, the spokesperson said, such as the misuse of powerful AI systems by “non-state actors and malicious entities”.
但澳大利亚战略政策研究所专注于中国人工智能的分析师亚历克斯·科尔维尔表示，新的合规要求可能会延迟或阻止中国人工智能公司与全球云服务提供商合作部署其最强大的开放权重模型，因为这将涉及将模型的“权重”输出中国境外。
Still, the new compliance requirements might delay or prevent Chinese AI companies from partnering with global cloud providers to deploy their most powerful open-weight models, as this would involve exporting the “weights” of the models outside China, said Alex Colville, an analyst at the Australian Strategic Policy Institute specialising in Chinese AI.
他表示：“我们仍然有很多事情不知道，”并指出，很多取决于计划最终版本中关键条款的表述方式，例如开源人工智能技术的“自动批准”在实践中意味着什么。
“There’s still a lot of things we don’t know,” he said, noting that much depends on how key provisions are worded in the finalised version of the plan, such as what “automatic approval” for open-sourcing AI technologies would mean in practice.
“到目前为止，中国方面一直以在人工智能监管方面尽可能灵活而自豪。如果未来六个月内该领域发生剧烈变化，他们只需调整出口管制要求以反映这种变化，”他说。
“The Chinese side has prided themselves on being as flexible as possible in AI regulation so far. If the field dramatically changes within the next six months, they can just change the export control requirements to reflect that,” he said.
当局周五表示，该草案计划的为期一个月的公众咨询期将持续至11月15日。
A month-long public consultation period for the draft plan would last until November 15, the authorities said on Friday.
上海德恒律师事务所高级合伙人尤云亭指出，这些新措施似乎是当局对美国科技巨头Meta Platforms今年早些时候终止收购中国人工智能代理开发商Manus事件的有针对性的反应。
You Yunting, senior partner at Shanghai DeBund Law Offices, noted that the new measures seemed to be a targeted reaction by the authorities to US tech giant Meta Platforms’ aborted acquisition of Chinese AI agent developer Manus earlier this year.
其中一项新提出的措施将要求在受限中国技术转让给境外所有者或控制方之前，必须先向商务部进行备案。
One of the new proposed measures would require prior registration with MOFCOM before restricted Chinese technologies could change hands to an overseas owner or controller.
该公司母公司蝴蝶效应（Butterfly Effect）最初于2022年在中国大陆成立，去年迁至新加坡，使该公司得以规避美国的技术出口和投资管制，为其被美国科技巨头以20亿美元收购铺平了道路。
Originally founded in mainland China in 2022, Manus’ parent company Butterfly Effect relocated to Singapore last year, allowing the firm to sidestep US technology export and investment controls, which paved the way for its $2 billion acquisition by the US tech giant.
然而，北京不久后以国家安全为由叫停了该交易。蝴蝶效应上个月宣布，已正式恢复独立运营。
However, Beijing called a halt to the transaction soon after, citing national security grounds. Manus announced last month that it had formally resumed independent operations.
您指出，国内人工智能公司出海的合规成本在短期内将上升，但出口管制可能为该国带来长期竞争优势。
You noted that compliance costs for domestic AI companies venturing abroad would rise in the short term, but the export controls could yield long-term competitive advantages for the country.
“鉴于该领域的巨额国内投资，实施合理管制是必要的[以保护这些投资]，”您说。
“Given the substantial domestic investments in the sector, implementing reasonable controls is necessary [to protect those investments],” You said.