US bars civilian agencies from using DeepSeek AI tools

The risk behind the restriction

A document attached to the notice says DeepSeek’s models pose several types of threat. One is that use by government employees or contractors could result in confidential information being transferred to China.

DeepSeek’s servers are in China. DHS says Chinese law gives the government “broad authority” to access private companies’ infrastructure.

The Department of Defense did not respond to AFP’s request for comment.

A policy decision amid competition

DeepSeek jolted the AI industry in January 2025 with a low-cost, high-performing chatbot, challenging assumptions about US dominance. The company introduced its long-awaited V4 model in April and updated a faster, more compact version in October.

The US has accused DeepSeek and other leading Chinese AI companies of systematically copying capabilities from advanced American models. The new restriction gives that rivalry a practical consequence for civilian government use, while leaving the military and intelligence services outside DHS’s reach.

In September, Chinese President Xi Jinping and US President Donald Trump met at the White House and agreed to establish a “communications channel” for discussing AI-related incidents. The issue is expected to come up again when they meet at a summit planned for November in Shenzhen, China.

I think the unresolved question is not only whether government users can keep sensitive data away from DeepSeek, but how the two sides will handle AI incidents while restricting access to each other’s technology. The planned channel now sits alongside a domestic ban—and neither, on its own, settles how much contact the competition can tolerate.