The Department of Homeland Security has ordered federal civilian agencies to remove DeepSeek products and third-party products “developed using DeepSeek implementations” from their computer systems, and barred future purchases of covered technology. The order, dated July 24, took effect on issuance and became public Monday after the 60-day window for agencies to request exceptions had already closed. The removal requirement reaches software sold under other companies’ names, beyond DeepSeek’s own app.

What Changed

AI-generated summary, reviewed by an editor. More on our AI guidelines.

What the order covers

The directive applies to civilian executive branch agencies, including Treasury, State and Justice. Defense is excluded, while national security systems fall under separate authorities.

DHS acted under the Federal Acquisition Supply Chain Security Act of 2018 after reviewing a Federal Acquisition Security Council recommendation and supporting evidence. The procurement exclusion names Hangzhou DeepSeek Artificial Intelligence Co., Ltd. and its affiliates.

That third-party language brings other vendors’ products and services within the removal requirement. DeepSeek’s open-weight models can run locally or through another hosting provider. Perplexity runs DeepSeek on its own U.S. servers. The version served through DeepSeek’s website is hosted in China.

DHS’s stated security concerns

Homeland Security Secretary Markwayne Mullin said DeepSeek’s models present “multiple threat vectors.” The order cites servers in China and Chinese laws permitting broad government access to private companies’ infrastructure.

The notice does not detail the evidence behind the council’s recommendation or set a removal deadline.

In the signed order appended to the notice, DHS also cited the collection of personally identifiable information during account creation and the potential to track users through digital fingerprints. Information entered by government employees or vendors could reach China and, combined with publicly available data, provide intelligence about Americans or government programs.

The department said the models “may contain malware or other security vulnerabilities,” citing open-source research that found critical vulnerabilities in DeepSeek’s iOS application.

DHS and DeepSeek did not immediately respond to requests for comment, and the Chinese embassy in Washington did not immediately comment.

A ban already in force

The exclusion and removal took effect immediately on the order’s issuance, unless an agency obtained an exception. Formal Federal Register publication is scheduled for Tuesday, following the notice’s filing for public inspection Monday.

Agencies seeking exceptions had to submit written requests within 60 days of issuance, with a compelling justification such as an effect on mission-critical functions and proposed safeguards. Measured from July 24, that window closed before the notice became public.

Until removal is complete, Mullin’s order encourages agencies to monitor network traffic and strengthen logging and controls against data leaving their systems.

Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.

Several agencies had already imposed restrictions. Commerce and the Navy banned DeepSeek on government devices, while the Pentagon’s Defense Information Systems Agency moved to block its chatbot in January 2025. New York, Tennessee and Virginia also restricted its use.

Timing

DeepSeek released an updated version of its V4.1-Flash model, V4.1-Flash-1017, on October 17, two days before the notice became public.

Last week China proposed adding advanced AI technologies to its export-control list. The draft covers 23 AI technologies, including large language models, agents, chip designs and computing clusters, and restricted items would include model weights and source code for products above certain performance thresholds.

Limited usage data

There has been no reported mass use of DeepSeek across the federal government. An Agriculture Department employee attempted to access the chatbot last year.

Mullin must review the order within a year of its issuance and annually thereafter.

Frequently Asked Questions

What does the DHS order require?

Federal civilian agencies must remove DeepSeek products, affiliate products and third-party products and services developed using DeepSeek implementations from their IT systems. The order also excludes Hangzhou DeepSeek Artificial Intelligence Co., Ltd. and its affiliates from federal civilian procurement.

Which agencies are covered?

Civilian executive branch agencies, including Treasury, State and Justice. Defense is excluded, and national security systems fall under separate authorities.

Why did DHS act against DeepSeek?

The order says DeepSeek's models present multiple threat vectors, citing servers in China, Chinese laws permitting broad government access, collection of personal information, and possible malware or security vulnerabilities. The notice does not detail the evidence behind the recommendation.

When did the ban take effect?

The order is dated July 24 and took effect on issuance. It was filed for public inspection on October 19, with Federal Register publication scheduled for October 20. The 60-day window for agencies to request exceptions had closed by then.

Does the order cover DeepSeek models hosted in the U.S.?

The removal text covers third-party products developed using DeepSeek implementations. DeepSeek's open-weight models can run locally or with other hosts, such as Perplexity's U.S. servers, while DeepSeek's own website runs on servers in China.

AI-generated summary, reviewed by an editor. More on our AI guidelines.

Germany Targets DeepSeek App Ban Over China Data Concerns
Germany demands Apple and Google remove Chinese AI app DeepSeek from their stores, claiming it illegally ships user data to China. The move reflects growing global tensions over AI privacy and data sovereignty.
DeepSeek Froze AI Prices at a Level Western Labs Cannot Matc
DeepSeek removed the strikethroughs from its V4 Pro pricing page Saturday. The 75 percent discount is now permanent. Output tokens cost $0.87 per million — 34.5 times less than GPT-5.5. The gap runs on Huawei chips and widens with every Ascend 950 that ships.
Anthropic Says Alibaba Ran 28.8 Million Claude Exchanges
Anthropic says Alibaba-linked Qwen operators ran 28.8 million Claude exchanges through nearly 25,000 fake accounts. The claim lands as lawmakers weigh sanctions for model extraction, Alibaba fights a Pentagon label and Anthropic contests U.S. controls on its own models.
AI News

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai