US Orders Agencies Off DeepSeek AI Over China Data Fears

The US Department of Homeland Security ordered civilian government agencies on Monday to remove AI tools made by China's DeepSeek from their IT devices. The ban also covers tools built by third parties using DeepSeek's technology.

"DeepSeek's models present multiple threat vectors," said a document attached to the DHS notice. The department warned that use of the tools "may result in the transfer of sensitive data to [China] if used by US government employees or vendors."

Why did the DHS act now?

The agency pointed to where DeepSeek runs its infrastructure. Its servers sit in China, where the legal system gives the government "extensive authority" to access private company infrastructure, the department said. That access, combined with the sensitivity of government work, drove the decision.

The directive lands as Washington and Beijing push hard for dominance in AI. DHS can direct civilian agencies but has no authority over the military or intelligence services. The US Department of Defense did not respond to an AFP request for comment on whether it uses DeepSeek products.

DeepSeek has been contacted for comment.

What is DeepSeek?

One of China's top AI labs, DeepSeek shook up the industry in January 2025 with a low-cost, high-performance chatbot. The release challenged assumptions of US dominance in the field and forced rivals to reckon with a cheaper path to frontier-level performance.

The firm released its long-awaited V4 model in April. Earlier in October, it shipped an updated, faster and more lightweight version of V4. The lab has kept a steady release cadence since its breakout moment at the start of the year.

What does the ban change?

The DHS notice binds civilian agencies to strip DeepSeek products — and derivative tools from third parties — off government IT devices. That scope matters. It doesn't just block DeepSeek's own apps; it catches any vendor product that quietly runs on DeepSeek models underneath.

The US has accused DeepSeek and other leading Chinese AI firms of systematically stealing the capabilities of cutting-edge models developed by American rivals. Monday's directive fits that broader posture: keeping Chinese-built AI out of federal systems while the accusations over model theft play out.

Does diplomacy still have a lane?

Competition hasn't shut down communication between the two governments. When Xi Jinping and Donald Trump met at the White House in September, China and the US agreed to set up a "communication channel" for AI incidents.

The topic is expected to come up again. Xi and Trump are scheduled to meet at a summit in November in the Chinese city of Shenzhen, where AI will likely be on the agenda despite the hardening US policy line.

For vendors selling AI tools into federal agencies, the message is blunt: prove what's under the hood, or lose the contract.