It started Monday with a dry government notice in the Federal Register. By Tuesday morning, Beijing was accusing Washington of politicizing technology, and every federal civilian agency in America had a new item on its to-do list: find every trace of DeepSeek in its computers and rip it out.
This is not a ban on you. You can keep using DeepSeek on your phone tonight if you want. But if you work for the federal government, Washington just told your IT department that a Chinese chatbot is a national security problem, and it wants it gone.
And honestly? The scariest part of this order is not what it bans. It is how much it covers that nobody can clearly identify.
What the order actually says
On Monday, the Department of Homeland Security issued a formal order under the Federal Acquisition Supply Chain Security Act of 2018. The directive tells federal civilian agencies to do two things: stop buying DeepSeek products and services, and remove every DeepSeek implementation already sitting in their IT systems.
Here is the part that matters. The order does not stop at DeepSeek’s own chat app or its branded tools. It explicitly covers third-party products and services that were built using DeepSeek’s models. If some vendor sold the government a customer-service tool that quietly runs on DeepSeek under the hood, that tool has to go too.
Agencies can request exceptions, but the default position is now clear: DeepSeek is out of federal civilian systems, period. The department also advised agencies to keep a closer eye on their networks while the removal happens, watching for anything unusual as the banned tech is phased out.
Why Washington says DeepSeek is dangerous
The DHS notice did not mince words. It said DeepSeek’s models present “multiple threat vectors” and warned that using them could let sensitive information end up in the hands of the People’s Republic of China.
The specific concerns break down like this:
- DHS says DeepSeek’s servers are in China, and Chinese law gives the government broad authority to access private company infrastructure.
- The order claims DeepSeek’s account signup process collected personally identifiable information and may track and digitally fingerprint users.
- Combined with publicly available data, that information could provide “actionable intelligence on US persons and government programs,” the notice said.
- And in a line that should make any IT director nervous, DHS warned the models themselves “may contain malware or other security vulnerabilities,” noting that open-source security research had already found critical vulnerabilities in DeepSeek’s iOS app.
This is not the first squeeze. A law passed in December 2025 had already targeted DeepSeek’s use in intelligence and national security systems. This week’s order extends the walls to federal civilian agencies, the much bigger, messier world of regular government IT.
Beijing fired back within hours
China did not wait long. On Tuesday, Foreign Ministry spokesperson Mao Ning criticized the move, saying Beijing opposes “overstretching the concept of national security and politicizing technology and trade issues.” She added that China will “resolutely safeguard the legitimate rights and interests of Chinese enterprises.”
It is worth being honest here about what Beijing’s complaint does not prove. Washington’s stated security concerns are claims about risk, not courtroom-level proof that DeepSeek has handed over classified data. The order itself was issued after a review of a Federal Acquisition Security Council recommendation, but the notice did not publish the underlying evidence.
At the same time, this fight is happening in a week when the pressure is running in both directions. Bloomberg reported last week that China is considering export restrictions on 23 AI-related technologies, including chip designs and certain advanced AI models. The two countries are not just trading chatbot complaints. They are actively fencing off each other’s AI industries.
The messy part: nobody knows which products count
Here is where this order gets genuinely complicated. The DHS notice requires agencies to remove “all DeepSeek and DeepSeek-affiliate products, and third-party products and services developed using DeepSeek implementations.” But it publishes no list of those third-party products.
Think about what that means in practice. A federal employee might use a writing assistant, a code helper, or a data tool whose vendor switched to DeepSeek’s open-weight models last year because they were cheap and good. The employee has no idea. The IT team might not know either. As The Register pointed out, it is not even clear whether “developed using” means a product that still queries DeepSeek models through an API today, or something that was simply built with DeepSeek’s help and now runs on its own servers.
Then there is the AI gateway problem. Services that route your request to whatever model fits best could silently send a federal user’s prompt to DeepSeek without anyone choosing it. Agencies now have to figure out not just which products they bought, but which models actually answer the queries inside them.
Oh, and two notable names were absent from the order entirely: Moonshot AI’s Kimi and Alibaba’s Qwen, two other major Chinese open-weight models, were not named. Make of that what you will.
What this means for you
If you do not work for a federal agency, nothing in this order touches your phone or your laptop. Businesses and individuals are not prohibited from using DeepSeek. The order is strictly about government IT systems and government procurement.
But zoom out a little and the direction of travel is unmistakable. Washington treated AI models like consumer software two years ago. Now it treats them like supply chain equipment: every component’s country’s origin matters, and the wrong origin gets your product banned from government contracts. DeepSeek is the first big AI name to get this treatment across the whole civilian government. It will not be the last.
And there is a deeper irony here that nobody in Washington wants to dwell on. DeepSeek burst onto the scene last year with cheaper open-weight models that briefly wiped about a trillion dollars off US tech stocks. The models were so cheap and capable that American companies built on top of them, which is exactly why this removal order is such a headache to execute. The technology Washington now wants gone is woven into the products its own agencies bought.
The race between the US and China in AI has narrowed to a matter of months on some measures, according to earlier Wall Street Journal reporting. Orders like this week’s will not widen that gap on their own. They just make sure that when the next breakthrough lands, the servers it runs on are on the right side of an ocean.
Your move, Beijing.
Sources: TechCrunch; The Wall Street Journal; The Register.
Read more in our Technology section, including how OpenAI’s own models went rogue last week and the White House mandate forcing AI companies to report every agent mistake.
