←新闻详情

The Elec更新于 10月21日 22:36

美国土安全部要求民用联邦机构移除 DeepSeek 模型与相关服务

US Bans DeepSeek Across Civilian Federal Agencies < IT·Gaming < 기사본문 - The Elec Inc.

AI 导读

美国国土安全部(DHS)宣布措施,要求民用联邦机构从政府 IT 系统中移除 DeepSeek 的模型和产品,以及集成了这些模型的第三方服务,并禁止机构采用基于 DeepSeek 产品开发的第三方服务,这实际上切断了政府机构通过 Microsoft、AWS 等平台访问 DeepSeek 的途径。DHS 给出的理由是敏感信息可能流向中国、模型可能被用于生成针对美国人的情报,以及模型本身可能含恶意代码或漏洞。报道同时提到,据 OpenRouter 流量分析,中国模型处理的 token 占比一度达到 58%,7 月第一周短暂升至 63%,而美国模型份额从一年前的 70% 降至 30% 区间。

推荐理由

AI 评分 70

报道梳理了美国土安全部对 DeepSeek 的禁用范围,并给出中国模型在美企业 token 占比升至 58% 的对照数据。

主题

标签

正文

当前显示原文,中文全文译文尚未就绪。

가

The body content of the article will be changed to this letter size.

스크롤 이동 상태바

The US Department of Homeland Security (DHS) has announced measures requiring civilian federal agencies to remove Chinese AI company DeepSeek’s models and products, as well as third-party services incorporating them, from government IT systems.

The department has also prohibited agencies from adopting third-party services developed using DeepSeek products, including its AI models. The restrictions effectively prevent government agencies from accessing DeepSeek through platforms such as Microsoft and Amazon Web Services (AWS).

Civilian federal agencies comprise executive branch departments and agencies outside the military and intelligence community. The measure therefore amounts to a ban across much of the US executive branch.

Military and intelligence IT systems are managed separately by the Department of Defense and the Intelligence Community (IC). The US Navy and Defense Department had already blocked access to DeepSeek.

DHS cited multiple security threats associated with DeepSeek models, including the possibility that sensitive information could be transferred to China if the products were used by US government employees or contractors.

The department also warned that government use of DeepSeek could generate intelligence exploitable against Americans and US government agencies, and that the models themselves might contain malicious code or other security vulnerabilities.

According to Reuters, the Defense Information Systems Agency (DISA) and NASA were among the first to block DeepSeek access in January 2025. The US Congress subsequently prohibited installation of DeepSeek on official devices used by congressional offices.

The Commerce Department, Navy and several state governments, including Virginia, Texas and New York, have also imposed restrictions on DeepSeek use on government devices.

DeepSeek is currently the only Chinese AI model subject to the latest prohibition, but the same executive order could potentially be extended to other Chinese models, including those developed by Moonshot AI and Alibaba.

However, a nationwide ban on such models could prove difficult because they are open-weight models. Experts have argued that a blanket prohibition on open-weight models could conflict with the First Amendment to the US Constitution.

Meanwhile, Chinese AI models are gaining traction among US businesses as companies seek to reduce AI token costs.

Bloomberg and CNBC reported in late July, citing traffic analysis from AI gateway service OpenRouter, that Chinese models accounted for a record 58% of tokens processed, briefly reaching 63% in the first week of July.

According to the analysis, the share of traffic handled by US-developed models fell from 70% a year earlier to the 30% range.

Hyun-Seon, Park hspark@thelec.kr

My comment

来源

The Elec