DHS orders federal agencies to purge DeepSeek AI tools from their networks

The U.S. Department of Homeland Security has ordered federal civilian agencies to stop buying and to remove products built on DeepSeek, the Chinese AI startup, from government computer systems. The directive cites national security concerns about how the company handles user data.

DHS said sensitive information from U.S. government users could become accessible to Chinese authorities under Chinese law. The order does not stop at DeepSeek’s own chatbot or app. It also covers third-party products and services built using DeepSeek’s underlying technology, though agencies can request exceptions.

The directive extends restrictions that were narrower until now. A law passed in December 2025 already barred DeepSeek’s use in national security systems run by U.S. intelligence agencies. The new order widens that ban to cover federal civilian agencies, applying to both new purchases and software already installed on existing systems.

China’s government pushed back. Foreign Ministry spokesperson Mao Ning said Tuesday that Beijing opposes “overstretching the concept of national security and politicizing technology and trade issues,” according to an official briefing. She added that China will safeguard the legitimate rights and interests of Chinese enterprises.

DeepSeek has drawn scrutiny from governments since January 2025, when its R1 reasoning model unsettled the AI industry and triggered a sharp sell-off in Nvidia shares. The company has since become a central figure in U.S.-China competition over artificial intelligence, offering low-cost, open-weight models that compete with products from OpenAI and Meta.

The scrutiny extends beyond data handling. A September report from the U.S. accused DeepSeek and five other Chinese AI companies, including Alibaba and Moonshot AI, of using large-scale model distillation, a technique that trains a new model on another model’s outputs, to extract information from American AI systems. OpenAI has told U.S. lawmakers it observed suspected attempts by DeepSeek to distill its models, and Anthropic has said it identified unauthorized attempts by DeepSeek and other Chinese AI labs to collect outputs from its Claude models.

The dispute is not one-sided. Bloomberg reported last week that China is considering export restrictions on 23 AI-related technologies, including chip designs and some advanced AI models, citing security and misuse risks of its own.

Agencies now face the task of identifying which systems run on DeepSeek technology, including tools built by third parties, and either removing them or applying for an exception. How strictly that exception process is enforced, and whether other governments follow Washington’s lead, will determine how far the order’s practical reach extends.