DHS Orders DeepSeek Purged From Federal Civilian Systems
3 min readThe US Department of Homeland Security has ordered every federal civilian agency to strip DeepSeek technology out of its IT systems and to stop buying anything built on it. The DeepSeek ban, dated late July, became public in a Federal Register notice on Monday, and Beijing answered within a day. It is the broadest action Washington has taken against a single Chinese AI company so far.
Background: a free model that went everywhere
DeepSeek, based in Hangzhou, shook the industry in early 2025 with open-weight models that matched American systems at a fraction of the training cost. Because the weights are free to download, the models spread quickly into third-party apps, cloud catalogs and internal tools. Microsoft listed the R1 model on Azure AI Foundry and GitHub Models in January 2025, and the US Navy barred its personnel from using DeepSeek that same month.
Until now the restrictions were piecemeal. Washington has also accused Chinese labs, DeepSeek among them, of large-scale distillation, meaning they trained on the outputs of frontier American models, a claim Beijing rejects.
What the DeepSeek ban requires
According to the Associated Press, agencies must remove DeepSeek and its affiliated products from their systems, remove third-party products and services developed with DeepSeek, and stop procuring any of it. That covers the chatbot, the downloadable weights and the API. DHS acted under the Federal Acquisition Supply Chain Security Act of 2018 on a recommendation from the Federal Acquisition Security Council, and CISA guidance says compliance is mandatory. Exceptions require a written case for “compelling mission-critical or national-security needs.”
The notice says DeepSeek’s models “present multiple threat vectors,” pointing to servers located in China and the possibility that Chinese authorities could reach sensitive US data. DeepSeek’s own privacy policy, updated in March, says prompts, uploads and chat history are stored in China. No removal deadline has been published, and it is unclear how many agency systems are affected.
Beijing’s response
On Tuesday, Chinese Foreign Ministry spokesperson Mao Ning said China “will resolutely safeguard the legitimate rights and interests of Chinese companies” and accused Washington of overstretching the concept of national security and politicizing technology and trade. DeepSeek did not comment.
Why it matters
The hard part is the third-party clause. An agency may never have installed DeepSeek yet still depend on a contractor tool that quietly calls a DeepSeek model, which is why CISA is telling agencies to audit their inventories and subcontractors. Expect vendors to be asked to certify exactly which models sit under their products.
The timing is pointed too. President Trump and President Xi are due to meet at the APEC summit in Shenzhen in mid-November with AI on the agenda, days after Beijing proposed export controls on its own frontier models. Both governments are now drawing lines around whose AI is allowed to run where. Watch for a compliance deadline, for allies to copy the order, and for how US cloud providers that still host DeepSeek weights respond.