.png "Global Insight")
The United States has ordered Chinese AI lab DeepSeek off its government's computers. The Department of Homeland Security has issued an exclusion and removal order directing every federal civilian agency to strip DeepSeek products from its IT systems — and the ban reaches further than the company's own app.
Here is what the order actually requires, why Washington says it had no choice, and the loophole that could make enforcement genuinely difficult.
What the ban covers
The notice sets out two actions. First, it excludes IT products and services developed by Hangzhou DeepSeek Artificial Intelligence Co. — including affiliated entities — from federal civilian agency procurements. Second, it requires the removal of all DeepSeek products, and third-party products and services built on DeepSeek's technology, from civilian agency IT systems.
That second clause is the wide one. DeepSeek releases its models as open weights, so they turn up inside other companies' tools as well as in DeepSeek's own app and website. As one analysis of the order notes, a tool "developed using DeepSeek implementations" is covered — software carrying someone else's brand but running on DeepSeek's models underneath is in scope. The order does not list which products it means.
The order was issued under the Federal Acquisition Supply Chain Security Act of 2018, following a recommendation from the Federal Acquisition Security Council. According to Reuters, the signed order is Exclusion and Removal Order #2026-01, publicly noticed on October 19 under Federal Register docket DHS-2026-0628.
Why DHS says DeepSeek is a threat
The department calls DeepSeek's models "multiple threat vectors." Its stated concerns: use by government employees or vendors "may result in the transfer of sensitive data to the PRC," could "produce actionable intelligence on US persons and government agencies," and the models could contain malware or other security vulnerabilities, given that Chinese cyber and intelligence laws permit state access to private company infrastructure.
The order applies to federal civilian executive-branch agencies — it does not cover the Pentagon or intelligence agencies' national-security systems, and it does not ban private businesses or individuals in the US from using DeepSeek. Some agencies, including the Commerce Department and the Navy, had already blocked DeepSeek on work devices; South Korea restricted the service on government computers back in February 2025.
The awkward timing
The ban landed days after DeepSeek released an updated cheap Flash model, which the company claims beats Anthropic's Claude Opus 5 at some coding tasks — a reminder that the lab Washington is walling off is still shipping competitive products. That tension sits at the heart of the story: the US wants DeepSeek's technology out of its government while American AI labs race the same company for capability leads, a race we covered in our piece on Anthropic's White House briefing.
The real enforcement test will be the third-party clause: auditing every software vendor for DeepSeek-derived models is a far harder job than uninstalling one app. Washington has drawn the line — now it has to find everything on the wrong side of it.
Is banning one AI lab enough when its open models are already everywhere? Follow Global Insight for clear-eyed coverage of the AI power struggle.