内容来源：https://www.wired.com/story/the-white-house-is-trying-to-figure-out-what-to-do-about-chinese-ai/

内容总结：

据多位知情人士透露，特朗普政府内部就如何应对中国人工智能（AI）模型快速崛起的局面存在严重分歧。这场争论主要分为两派：白宫方面主张对中国AI实施更严格的管控，因其可能很快匹敌美国最强大的模型；而商务部则认为这些限制措施不可行。

这一难题已成为美国政府AI政策的最新焦点。此前，中国月之暗面公司（Moonshot AI）发布了Kimi K3模型，其性能已逼近美国Anthropic和OpenAI的顶尖模型。知情人士称，近几日白宫高级官员讨论了一系列可能措施，旨在阻止中国AI实验室通过“模型蒸馏”技术（即利用美国模型训练中国模型）来开发自己的模型。

据悉，在Anthropic指控中国科技公司阿里巴巴发动了迄今“最大规模的蒸馏攻击”后，白宫预计将采取某种总统级行动，但目前尚不涉及行政命令。白宫发言人对相关报道不予置评。

白宫对蒸馏攻击尤为重视。周三，美方宣布月之暗面是通过蒸馏Anthropic的Claude Fable 5模型才开发出Kimi K3。财政部长斯科特·贝森特警告称，隐蔽的蒸馏行为可能导致制裁，并将其定性为“知识产权盗窃”。

尽管白宫内部提案流传已久，但据知情人士透露，目前尚未向商务部提交正式意见征询，而商务部自有主张。商务部在政府内部分歧中影响力日益上升，部分原因在于其下属的工业与安全局负责出口管制。

知情人士补充说，商务部长卢特尼克正在考虑制定激励措施，鼓励美国顶尖实验室发布开放权重模型来与中国抗衡，并已在最近几周与多家AI实验室负责人会谈。有消息称，卢特尼克似乎在监管中采取折中立场——他通过出口管制约束了Anthropic，但比凯恩克罗斯更为灵活。

本月早些时候，美国政府面临的最大AI议题是Anthropic的Claude Mythos和Fable 5模型，因为它们能发现政府系统和关键基础设施的漏洞。但随着中国所谓的“开放权重模型”崛起（这些模型几乎没有防止黑客入侵政府系统的安全措施，且可被任何人下载使用），AI领域的政策难题进一步升级。

中文翻译：

据多位熟悉讨论情况的人士透露，特朗普政府内部在如何应对中国领先人工智能模型迅速崛起的问题上存在分歧。这场争论大致分为两派：白宫方面推动对中国人工智能实施更严格的管控，认为其可能很快与美国最强大的模型并驾齐驱；而商务部则认为这些管控措施不可行。这一难题已成为政府在人工智能领域的最新冲突点。此前，中国月之暗面实验室于上周发布了其Kimi K3模型，该模型可与美国Anthropic和OpenAI的顶尖模型相抗衡。知情人士称，近几日，白宫高级官员讨论了可能采取的措施，以阻止中国人工智能实验室通过在美国模型上进行训练（即所谓的“蒸馏”技术）来开发自己的模型。预计白宫将在Anthropic上月指控中国科技公司阿里巴巴对其发动了“有史以来最大规模的已知蒸馏攻击”后采取某种总统行动。据其中一位知情人士透露，目前预计这一行动不会采取行政命令的形式。白宫发言人未就本文报道置评。白宫对蒸馏攻击尤为重视。其于周三宣布，月之暗面是通过蒸馏Anthropic的Claude Fable 5模型开发出Kimi K3的。财政部长斯科特·贝森特告诉福克斯商业频道，隐蔽的蒸馏行为可能导致制裁，并称其为“知识产权盗窃”。两位熟悉讨论情况的人士表示，尽管白宫内部正在传阅相关提案，但尚未向商务部发出正式意见征求，而商务部也有自己的想法。商务部在政府内部关于人工智能的分歧中，已日益成为有影响力的一方，部分原因在于其通过工业与安全局负责出口管制。消息人士补充称，卢特尼克已在考虑如何激励美国顶尖实验室创建自己的开放权重模型，以与中国抗衡，并在最近几周与多家人工智能实验室的负责人进行了交流。其中一位消息人士表示，卢特尼克似乎在监管上走中间路线——他对Anthropic实施了出口管制以使其就范——但比凯恩克罗斯更为灵活。本月早些时候，政府面临的最大人工智能问题是Anthropic的Claude Mythos和Fable 5模型，因为它们能够发现政府系统和关键基础设施中的漏洞。然而，随着中国所谓的“开放权重”模型（这类模型几乎没有或根本没有防止黑客攻击政府系统的防护措施，且任何人都可以下载和使用）的崛起，人工智能带来的政策难题只增不减。本文为雨果·洛威尔《内部循环》通讯专栏的一部分。点击此处阅读往期专栏。

英文来源：

The Trump administration is split over how to respond to the rapid rise of China’s leading AI models, according to multiple people familiar with deliberations.
The debate is broadly divided between parts of the White House, which has pushed for stricter controls on Chinese AI that may soon rival the most powerful US models, and the Commerce Department, which has viewed those restrictions as unworkable.
The conundrum has become the latest flash point for the administration over AI, after China’s Moonshot AI lab last week released its Kimi K3 model that rivals the top US models from Anthropic and OpenAI.
In recent days, senior White House officials have discussed possible measures to stop China’s AI labs from developing their models by training them on US models, a practice known as distillation, the people say.
The White House is expected to take some sort of presidential action after Anthropic last month alleged the Chinese tech company Alibaba had carried out the “largest known distillation attack” against it to date. This action is not currently expected to take the form of an executive order, according to one of the people.
A White House spokesperson did not comment on the reporting in this story.
The White House has been taking distillation attacks especially seriously. It announced on Wednesday that Moonshot had developed Kimi K3 by distilling Anthropic’s Claude Fable 5 model, and Treasury secretary Scott Bessent told Fox Business that covert distillation could lead to sanctions, describing it as “IP theft.”
While proposals have been making the rounds inside the White House, no formal request for input has been transmitted yet to the Commerce Department, which has its own ideas, according to two people familiar with deliberations.
The Commerce Department has increasingly emerged as an influential part of the intra-administration split over AI, in part because it oversees export controls through the Bureau of Industry and Security.
Lutnick has contemplated ways to create incentives for top US labs to create their own open weight models to counterbalance China, and has spoken with leaders at a number of AI labs in recent weeks, the people add.
Lutnick appears to be straddling a middle ground of regulation—he imposed export controls on Anthropic to bring them to heel—but has been more freewheeling than Cairncross, one of the people said.
Earlier this month, the biggest AI issue for the administration was Anthropic’s Claude Mythos and Fable 5 models, because of their ability to find vulnerabilities in government systems and critical infrastructure.
But policy headaches with AI have only escalated with the rise of so-called open weight models from China—models that have few to no safeguards to prevent hacking government systems and can be downloaded and used by anyone.
This is an edition of Hugo Lowell’s Inner Loop newsletter. Read previous newsletters here.
Comments
Back to top