WASHINGTON — The US Department of Homeland Security has told civilian federal agencies to purge software from Chinese artificial intelligence developer DeepSeek from their networks and to halt any new purchases, according to a notice made public on Monday.

The order reaches beyond DeepSeek's own chatbot and apps. It also covers third-party tools built on the company's AI models, The Next Web reported, and it shuts DeepSeek out of federal contracting. Until now, individual agencies decided for themselves whether to block such models, CyberScoop reported.

The notice was posted on the Federal Register website on Monday and is due to be formally published on Tuesday, CyberScoop reported. It applies to DeepSeek and its affiliates and bars agencies from buying any information technology products or services from the parent company, Hangzhou DeepSeek Artificial Intelligence.

The order relies on a 2018 supply chain security law that allows Washington to exclude suppliers it judges risky, and on a review by the Federal Acquisition Security Council. Agencies must leave DeepSeek out of future procurements and remove any of its products already in place. Engadget said the rule covers civilian bodies such as the Environmental Protection Agency and the Federal Communications Commission.

Homeland Security Secretary Markwayne Mullin signed the order months before it became public. Bloomberg reported that it is dated July 24, according to The Next Web, while CyberScoop said the notice includes a letter from Mullin ordering the exclusion on June 26. The Hill said the order was first issued in July.

In the order, Mullin said DeepSeek's models present "multiple threat vectors" to civilian agency networks. He pointed to Chinese national security laws that give Beijing wide powers to demand data from domestic companies, and said DeepSeek stores data on servers in China.

Mullin wrote that the prompt-and-response process used by the models "may result in the transfer of sensitive data" to the Chinese government, CyberScoop reported. DHS also said the models could "contain malware or other security vulnerabilities," according to Engadget.

"The transfer of the aforementioned information to the [People's Republic of China], either independently or in combination with publicly available data, could produce actionable intelligence on U.S. persons and/or government programs," Mullin wrote.

Some of the concerns DHS lists, such as mass data collection and the tracking of users' online fingerprints, could equally apply to American AI models, CyberScoop noted.

Agencies can seek an exception, but they must apply 60 days in advance and offer a "compelling justification," such as being unable to carry out a mission-critical function without the product. Until removal is complete, agencies are urged to watch traffic in and out of their networks, step up logging, tighten controls on data leaving their systems and flag unusual activity, according to the order.

DeepSeek did not comment, The Next Web said. The Chinese embassy in Washington did not immediately respond to a request for comment from Bloomberg. CyberScoop said it had asked DHS and the Cybersecurity and Infrastructure Security Agency for more details.

The order landed days after DeepSeek released its latest model, V4.1-Flash-1017, which the company described as "smarter, faster, more efficient," Engadget reported.

DeepSeek's cheaper models have begun to rival ChatGPT and Gemini, Engadget said. More than a year ago, US frontier AI companies and the federal government accused DeepSeek's makers of "systematically" distilling their models, according to CyberScoop.

Several arms of the government had already acted alone, The Next Web reported. In January 2025 a Pentagon IT agency cut off access to the DeepSeek chatbot, while the Navy and the Commerce Department barred it from work devices. The states of New York, Tennessee and Virginia imposed comparable limits.

The pressure on Chinese AI firms has built for months. OpenAI told US lawmakers in February that DeepSeek had used a technique called distillation, which trains a model on the outputs of another, to build its R1 chatbot, Bloomberg reported, according to The Next Web. US officials have accused Chinese labs of copying American models in this way.

Washington has stopped short of harsher measures. In June, the White House declined to put DeepSeek and more than 100 other Chinese companies on a US trade blacklist, even though a panel of officials had approved the step, according to The Next Web.

DeepSeek is the only Chinese AI developer named in the order. Other Chinese open-weight models, including Moonshot AI's Kimi and Alibaba's Qwen, have drawn similar concerns in Washington but are not covered, CyberScoop said.

The order is set to be formally published in the Federal Register on Tuesday, and any agency that wants to keep using a DeepSeek-based tool must file its request for an exception 60 days ahead.