The US has moved from China-related AI warnings to an actual order.
The Department of Homeland Security has directed all US civilian federal agencies to strip DeepSeek products out of their information technology systems, and has barred those agencies from buying any DeepSeek-linked products going forward.
The directive, published as a notice on Monday, invokes the Federal Acquisition Supply Chain Security Act of 2018, a law originally built to keep specific vendors out of government supply chains over national security risk. This is reportedly the first time it’s been used against an AI company, and it covers not just DeepSeek’s own products but any third-party service built using DeepSeek’s models.
DHS didn’t hold back on the reasoning. It said DeepSeek presents “multiple threat vectors,” and that its use by government employees or vendors “may result in the transfer of sensitive data to the PRC.” The department went further, warning that the models’ outputs “could produce actionable intelligence on US persons and/or government agencies,” and flagged the possibility that “the models themselves may contain malware or other security vulnerabilities.”
A July memo, only made public alongside Monday’s notice, shows DHS had already quietly told federal agencies to start monitoring data flowing in and out of government websites to watch for DeepSeek-related risk, well before this became a formal order.
This Has Been Building For A While
DeepSeek has had a target on its back in Washington since its R1 model rattled markets in January 2025. Individual agencies didn’t wait for a government-wide rule — the Department of Commerce and the Navy had already banned it on staff devices, and states including New York, Virginia, and Tennessee put similar restrictions in place on state government hardware.
Congress tried to get ahead of it too. The “No DeepSeek on Government Devices Act” was introduced in the House in February 2025, with a companion bill in the Senate, aiming to push OMB into mandating removal across federal tech. Neither bill actually became law — which is what makes this DHS order notable. It’s not Congress legislating a ban; it’s an executive agency using existing acquisition-security authority to do it directly, without waiting on the Hill.
It’s also not an isolated move. In July, the FCC voted to block sales of devices containing key hardware from Chinese firms it considers national security risks, and has floated barring US telecom carriers from interconnecting with Chinese telecom companies on the same list. The DeepSeek order fits into that same broader pattern of the US trying to wall off its government and critical infrastructure from Chinese tech, component by component. It likely hadn’t helped that a US govt website was discovered to be using China’s Qwen model, before it was hurriedly taken down.
The Timing Is Awkward For DeepSeek
The ban lands just two days after DeepSeek shipped an update to its V4-Flash model, which reportedly improved benchmark scores and cut down on task runtimes without raising API prices. The company has kept shipping competitive, cheap models through all of this scrutiny, which is precisely what’s made Washington uneasy — DeepSeek’s low-cost approach has been chipping away at the idea that the US has a clean lead in frontier AI.
For now, the order is specific: it covers federal civilian agency IT systems and procurement, not general use by American companies or individuals. DeepSeek remains freely downloadable in the US for anyone outside the federal government. But as agencies keep tightening what AI tools and vendors they’ll touch, the line between “banned for government use” and “banned, full stop” is one that’s been getting thinner with each new order.