Federal civilian agencies in the United States must stop using AI technology from Chinese lab DeepSeek. The US Department of Homeland Security’s order requires agencies to remove existing DeepSeek products from their IT systems and bars them from buying new ones.
The scope extends beyond products directly from DeepSeek: it also covers affiliated companies and third-party services built with DeepSeek implementations. But the notice leaves practical questions unanswered, including when agencies must comply and which services count as being “built with” the technology.
What the order requires
Agencies face two direct requirements: remove DeepSeek products already in their systems and avoid new purchases. The order also reaches products from affiliated companies, as well as third-party services built using DeepSeek implementations.
That broader language could matter for services that connect users to different AI models. The article says it remains unclear whether AI gateways such as OpenRouter fall under the ban. Without a clearer definition of “built with,” agencies may have difficulty determining which products and services are covered.
Key details remain unspecified
The department did not set a deadline for removing the technology. It also did not name the specific security concerns behind the decision. As a result, the order communicates what agencies must do, but the source article reports that it does not explain the rationale or timeline in detail.
Those gaps leave implementation questions open. Agencies may need to assess whether products in their IT systems use DeepSeek directly, come from an affiliated company, or rely on a third-party implementation. The article does not describe how the department will interpret those categories or how agencies should resolve uncertain cases.
A procurement security basis
The order is based on a 2018 US law covering supply chain security in federal procurement. That places the restriction in the context of federal purchasing and the security of technology used by government agencies.
The article does not identify which particular supply chain risks informed this decision. It also does not say when the department will provide more guidance. For agencies, the immediate task is to identify covered technology and prepare to remove it, while the precise boundaries and timing remain unsettled.
Part of a wider debate over Chinese AI models
The move follows earlier consideration by the US government of possible steps targeting Chinese open-weight models. According to the New York Times, options discussed over the summer included sanctions, a trade blacklist, and rules limiting which models federal agencies could use.
The government pulled back for now after Nvidia, Microsoft, Google, and Meta objected. OpenAI and Anthropic had lobbied for restrictions. The article presents those discussions as background to the current order, while leaving unclear how the broader debate will develop.
For now, the ban’s practical effect is focused on federal civilian agencies and DeepSeek-related technology. Its reach may depend on how the department interprets affiliated products and third-party services, questions the notice did not settle.