US bars civilian federal agencies from using DeepSeek AI, citing data risks

The order

The US Department of Homeland Security has ordered civilian federal agencies to take DeepSeek products off their IT systems and stop buying them. The notice was posted in the Federal Register on Monday, 19 October, but the order itself is dated 24 July and was signed by Homeland Security Secretary Markwayne Mullin. The restriction took effect at the end of July, so agencies were already bound by it before the public notice appeared. The order also covers tools from other companies that are built on DeepSeek's AI models, and it bars the Chinese company from taking part in federal contracts. It was issued under a 2018 supply chain security law that lets the government cut risky suppliers out of its purchasing.

Scope and stated risks

The order applies to civilian agencies, so it does not reach the military or the intelligence services, and the Department of Defense is excluded. Civilian bodies such as the Environmental Protection Agency and the Federal Communications Commission would have to drop DeepSeek products and remove existing ones. The department described the risk as the possible transfer of sensitive data belonging to US government employees and vendors to the Chinese government. It also said the company collects large amounts of user data that could be passed to Beijing under Chinese security laws, and that the models pose multiple threat vectors and may hide malware or security flaws. The notice itself does not spell out the exact security concerns, so most of the stated reasoning stays general.

Earlier restrictions and pressure

Several parts of government had restricted DeepSeek before this order. A Pentagon IT agency blocked its chatbot in January 2025, and the Navy banned its use for internal purposes the same month. The Navy and the Commerce Department have also banned the tools on work devices, and New York, Tennessee and Virginia have taken similar steps. The Czech government banned the use of any DeepSeek products in state administration in 2025, acting on advice from the National Office for Cyber and Information Security (NÚKIB).

In June, the White House declined to add DeepSeek and more than 100 other Chinese firms to a US trade blacklist, even though a panel of officials had cleared them for listing.

Copying accusations and the AI race

The rivalry behind the order is shaped by claims of copying. In February, OpenAI told US lawmakers that DeepSeek had used distillation, a method in which a smaller model learns from a larger one, to build its R1 chatbot. Anthropic and OpenAI, both of which are expected to list on public stock markets in the coming months, have also said that some Chinese AI firms, DeepSeek among them, essentially copied their models. White House technology lead Michael Kratsios wrote in an internal memo earlier this year that American AI firms were being maliciously exploited, mainly by rivals in China. Beijing has rejected the accusations, and the Chinese embassy in Washington did not immediately respond to a request for comment. President Donald Trump said on Monday that the US is far ahead of China in what he calls superintelligence, and he opposes state regulation of AI because it would protect that lead.

Adoption and open questions

DeepSeek has grown quickly since its first public release in 2025, and its free chatbot app ranks among the most downloaded apps in Apple's App Store. US companies including Snowflake, Vercel and OpenRouter build DeepSeek models into services they offer customers. Its latest model, V4.1-Flash-1017, appeared a couple of days before the order, and the company says it is smarter, faster and more efficient. Startups have turned to open models because using DeepSeek costs a fraction of what closed models from Google, Anthropic and OpenAI cost. DeepSeek did not comment on the order, and the Chinese embassy had not responded when the reporting was published.