DHS orders DeepSeek removed from civilian agency systems, Reuters reports

The US Homeland Security Department (DHS) is ordering civilian agencies to remove DeepSeek products, and third-party services built on the Chinese AI model, from their IT systems, Reuters reported on October 19, citing a notice published that day.

We have not seen the notice itself. Everything below is as reported by Reuters, and we could not confirm it against a DHS document.

A removal order plus a purchasing ban

According to Reuters, the order has two parts. Agencies must take DeepSeek products out of their systems, and they may no longer acquire DeepSeek products or third-party services developed using DeepSeek. The second clause matters because it reaches beyond DeepSeek’s own app. Software from other vendors that is built on the model falls inside the reported scope.

Reuters says DHS described DeepSeek models as posing “multiple threat vectors”. The department reportedly warned that using the product “may result in the transfer of sensitive data to the PRC if used by U.S. government employees or vendors.” Per the report, DHS also said agency use could yield actionable intelligence on US persons or agencies, and that the models might contain malware or other security vulnerabilities. These are DHS’s stated concerns as relayed by Reuters. The report does not describe any specific incident.

Reuters also reported that a July DHS memo, made public on Monday, told federal agencies to monitor traffic flowing in and out of government websites to address possible DeepSeek risks. The Chinese Embassy in Washington had not immediately commented, according to the report.

Who is affected, and what is still unclear

The order, as described, covers US civilian agency IT systems. Individual users, private companies and state governments are not mentioned in the Reuters account. DeepSeek’s models remain available to the public, and nothing in the report says otherwise.

The coverage we could read leaves several practical questions open: how long agencies have to finish the removal, how they will find DeepSeek-based components inside vendor software, and whether contractors must certify that they are clean. Those details would likely sit in the notice, which we could not review.

The move extends earlier, narrower steps. Reuters notes that the Department of Commerce and the Navy had already banned DeepSeek on government devices, and that New York, Tennessee and Virginia took similar action. The new order is broader because it applies across civilian agencies and adds a purchasing ban.

Signals to watch after the DHS order

The first thing to watch is whether the removal order comes with a deadline and a reporting requirement. Without one, compliance will be hard to measure from outside.

A second sign is how far the “built using DeepSeek” language reaches. Many AI products wrap open models under their own brand. If agencies must trace model origins through vendor stacks, software suppliers to government may have to disclose which models sit underneath their products.

Third, the order fits a wider pattern that Reuters describes. The Federal Communications Commission is due to vote on October 29 on barring all Chinese labs from testing electronic devices for use in the US. Reuters notes that 82 percent of US electronics were tested in China last year. The FCC also banned imports of new models of Chinese drones and consumer routers in December, and in July voted to bar sales of US devices containing key hardware from Chinese companies deemed national security risks.

There is also a legal backdrop. Under the Federal Acquisition Supply Chain Security Act, the Secretary of Homeland Security can issue removal and exclusion orders for civilian agencies. Law firm Morgan Lewis noted that the first public order under that law came from the Office of the Director of National Intelligence in September 2025, and covered Acronis AG. Reuters does not say which authority DHS used for the DeepSeek order, so the link is unconfirmed.

For readers outside government, the direct effect is small for now. For vendors selling AI tools to US agencies, the signal is clearer: models of Chinese origin are becoming a procurement risk, whatever the brand on the product.