US Orders Federal Agencies to Remove DeepSeek as Company Open-Sources Its AI Serving Tools

US bans DeepSeek from federal systems and targets third-party products

The US government has ordered federal agencies to remove DeepSeek technology from their systems, and it extended the restriction to third-party products built using the Chinese company's AI models.

The Department of Homeland Security (DHS) published the directive on Oct. 20. It requires federal civilian agencies to remove DeepSeek products from their information technology systems and blocks future purchases.

The order also covers software and services made by other companies using DeepSeek implementations. That means American technology contractors that have built DeepSeek models into their products could be affected.

Key details of the directive

• DHS issued the directive on Oct. 20, ordering federal civilian agencies to eliminate DeepSeek products from their IT systems.


• Homeland Security Secretary Markwayne Mullin signed the order on July 24 under the Federal Acquisition Supply Chain Security Act of 2018.


• Agencies were told to start complying immediately and had a 60-day window to request exceptions.


• Restrictions also apply to other companies' products and services that use DeepSeek implementations.

What DHS says about the security concerns

DHS cited national security concerns about DeepSeek's operations in China. The department said government employees could enter sensitive information that might be sent to servers in China.

DHS also warned the technology could contain security vulnerabilities and that data collected through its services could reveal information about US government activities.

The move follows earlier US actions against DeepSeek by the Commerce Department, the US Navy, and several state governments.

DeepSeek releases its serving framework as open source

On the same day the directive was published, Oct. 20, DeepSeek said it had open-sourced DeepInfer, the inference framework that runs its large-scale AI services. Inference is the process of an AI model generating a response to a request, and a framework is the software that manages that work.

The release includes computing kernels, distributed inference tools, speculative decoding, and processing tools meant to improve throughput and lower the computing resources needed to generate AI responses.

DeepSeek also released deepseek-recipe, a set of software components that handle token encoding, prompt formatting, and API request processing. These tools let independent developers reproduce key parts of the company's API behavior when running its models.

Company engineer Cui Tianyi said: "We are releasing DeepInfer, deepseek-recipe, and other vital components for serving DeepSeek models," adding that the goal was to help more businesses and organizations worldwide deploy and profit from the technology.

What the release contains and who can use it

• DeepInfer supports DeepSeek-V4.1-Flash, a 552-billion-parameter model introduced in September that activates only a fraction of its parameters for each task.


• DeepSeek said the model's architecture needs substantially less memory and storage for inference caching, so providers can handle more requests with fewer computing resources.


• DeepInfer is distributed under the Apache 2.0 license, which allows commercial modification and redistribution.


• It supports Nvidia SM100 graphics processors and Huawei Ascend 950 chips, covering both American and Chinese hardware.


• The first release offers offline inference only. Operators must still supply parts such as cache management and load balancing to build a full production service.

Reaction from China

China condemned the US decision, saying Washington was politicizing technology and undermining commercial cooperation between the world's two largest economies, according to the source material.

What is confirmed

• DHS published the directive on Oct. 20, requiring federal civilian agencies to remove DeepSeek products from their systems and barring new purchases.


• The order, signed by Markwayne Mullin on July 24, covers third-party software and services that use DeepSeek implementations.


• DeepSeek announced the open-source release of DeepInfer and deepseek-recipe on Oct. 20.


• DeepInfer is released under the Apache 2.0 license and supports Nvidia SM100 and Huawei Ascend 950 chips.


• China publicly criticized the US decision.

What is still unclear

The source material does not name the contractors affected by the directive or say how many products use DeepSeek implementations. It also does not state whether any company has sought an exception or replaced DeepSeek components.

The source notes that hosting DeepSeek technology on private infrastructure would not automatically exempt a product from the federal order, because the order expressly covers third-party implementations. It is not clear how that rule will be applied in practice.

Why this matters

The directive adds compliance work for federal technology suppliers. They must check whether products using DeepSeek implementations can still serve government customers. Affected contractors may need to swap out those components or seek exemptions, which could send business to competing AI providers, according to the source material.

DeepSeek's open-source release lets cloud providers and businesses outside the US federal market run DeepSeek models on their own infrastructure, adjust the technology, and potentially lower AI operating costs. The release could widen DeepSeek's reach globally, but the US restrictions limit its adoption among federal contractors and technology suppliers.

Sources

• CryptoSlate