Back to Business & Economics

Washington orders federal agencies to pull DeepSeek out of government systems

The U.S. Department of Homeland Security ordered federal civilian agencies to stop purchasing products and services from DeepSeek and to remove the Chinese AI startup from government IT systems. The order, published on Tuesday, cites national security concerns tied to the company's operations in China. It lands on the civilian side of the government, the departments and agencies that run day-to-day federal operations.

The concern on paper is data. The order says DHS is worried about how DeepSeek handles user data and warns that sensitive information from U.S. government users could potentially be accessible to Chinese authorities under Chinese law.

The ban is broader than a product ban. It also covers third-party products and services built using DeepSeek technology. In practice, that means an agency cannot keep a SaaS tool or a packaged product if it runs on DeepSeek models underneath. Agencies can request exceptions.

The second layer of restrictions

Washington had already restricted DeepSeek inside parts of the government. A law passed in December 2025 targeted its use in intelligence and national security systems. The new order extends the restriction to civilian agencies, and it reaches both new purchases and systems that are already installed. The December law touched a narrow slice of the federal machine. The new order reaches procurement at a far larger scale, because civilian agencies are where most federal software buying happens.

The scrutiny started in January 2025, when DeepSeek's R1 reasoning model shook the AI industry and triggered a sharp sell-off in Nvidia shares. Since then, the lab has faced scrutiny from governments and regulators around the world.

The distillation accusations

The order lands in the middle of a string of accusations against DeepSeek and other Chinese labs. A September report said U.S. authorities accused DeepSeek and five other Chinese AI companies, including Alibaba and Moonshot AI, of using large-scale model distillation to extract information from American AI systems. Distillation is the practice of using one model's outputs to train another. In plain terms, a lab can improve its own models by querying a rival's and learning from the answers, without ever seeing the rival's weights or training data.

The major labs have filed their own complaints. Earlier this year, OpenAI told U.S. lawmakers it had observed suspected attempts by DeepSeek to distill its models. Anthropic said it identified unauthorized attempts by DeepSeek and other Chinese AI labs to collect outputs from Claude.

China answered the order the same day. Mao Ning, a Foreign Ministry spokesperson, said Beijing opposes overstretching the concept of national security and politicizing technology and trade issues, and that it will resolutely safeguard the legitimate rights and interests of Chinese enterprises, according to an official briefing.

The restrictions are not one-way. Bloomberg reported last week that China is considering export limits on 23 AI-related technologies, including chip designs and certain advanced AI models, citing security and misuse risks.

For anyone selling into the U.S. federal government, the order is a procurement rule: software built on DeepSeek technology is out of scope unless an agency wins an exception. The exception process is the only off-ramp, and the call sits with each agency. For teams choosing models, it adds provenance to the compliance checklist. A model from a Chinese lab now carries a different risk profile inside government systems than it does outside them.

Source: techcrunch.com

More news

Back to Business & Economics

Have something in mind?

Start a conversation