The US government has told its civilian agencies to stop using DeepSeek. The Department of Homeland Security published a notice on Monday ordering the Chinese AI model removed from federal civilian IT systems, and barring those agencies from buying it in the future. The order also reaches products from other companies that are built on top of DeepSeek, which is the part with the biggest practical impact.
It's an important step, but it's narrower than some headlines suggest. This isn't a nationwide ban, and ordinary users and businesses aren't affected. Here's what the order says, why DHS issued it, and what it leaves out.
What the order actually says
DHS directed civilian federal agencies to take DeepSeek products out of their systems. It also banned those agencies from acquiring DeepSeek products or any third-party services developed using the model. That second part matters. Without it, an agency could drop the DeepSeek app but keep using a vendor's software that quietly runs DeepSeek behind the scenes, and the ban would mean very little.
Alongside the notice, DHS made public a memo from July. In it, the department told agencies to watch the traffic flowing in and out of government websites, so they could spot risks tied to DeepSeek. In other words, the department was already tracking the issue months before the formal ban arrived.
One report links the order to a 2018 federal law on supply-chain security in government purchasing. If you mention that in your own article, check it against the DHS notice first, since I could only find it in a single source.
Why DHS says it's worried
The department's reasoning has three parts. The first is data. DHS said using DeepSeek could lead to sensitive information reaching China if government employees or contractors use it. The second is intelligence gathering. The department warned that the product could produce useful intelligence about US citizens or government agencies. The third is the technology itself, since DHS said the models could carry malware or other security weaknesses. It summed the risks up as "multiple threat vectors."
Those are serious claims, but it's worth being precise about what they are. They describe risks, not confirmed incidents. The available reporting doesn't show that any sensitive US government data was actually sent to China through DeepSeek. The ban is a precaution based on what could happen, which is a common approach in national security procurement, but it isn't the same as a finding that a breach occurred.
What the ban doesn't do
This is where many readers get confused, so it's worth being clear.
- It doesn't cover the public. The order applies to civilian federal agencies, their employees and their vendors. Nothing in the reporting says regular people or private companies can no longer use DeepSeek in the US.
- It doesn't touch DeepSeek's wider business. Its consumer product and its customers outside the US aren't directly affected, though losing a large and security-conscious buyer is still a hit.
- It isn't an export control. One report says the order doesn't, for now, add DeepSeek to the Commerce Department's Entity List or bring wider trade restrictions. That same report mentions an unconfirmed internal approval in June for a listing, so treat that as something to watch, not as settled.
- It doesn't cover everyone in government. The notice is aimed at civilian agencies, so check separately how defense and intelligence bodies handle the model.
This didn't come out of nowhere
DeepSeek is a Chinese AI lab that drew global attention in early 2025 with models that performed well at a low cost. Washington's concern started almost immediately. Within weeks, the Commerce Department and the US Navy had told staff not to use it, and Congressional offices were warned against installing it on official devices.
In February 2025, two House members, Republican Darin LaHood and Democrat Josh Gottheimer, introduced the No DeepSeek on Government Devices Act, a bipartisan bill modeled on earlier restrictions on TikTok. They later urged state governors to adopt similar bans. Several states did. New York, Tennessee and Virginia are among those that have already restricted it, and reporting from earlier this year also lists Texas.
It wasn't only the US. Governments in South Korea, Australia and Taiwan blocked their employees from using the app, and an Indian ministry told officials to avoid AI tools like DeepSeek and ChatGPT for official work. The federal order is the latest and broadest US step in a pattern that has been building for almost two years.
The wider push against Chinese technology
The timing sits inside a bigger story. The Federal Communications Commission is scheduled to vote on October 29 on barring all Chinese labs from testing electronic devices such as phones, cameras and computers for use in the US. The scale matters here: last year, 82% of US electronics were tested in China. In July, the FCC also voted to stop sales of devices in the US that contain key hardware from Chinese companies it considers national security risks.
Taken together, these steps show Washington moving from case-by-case reactions to a broader strategy of limiting Chinese technology where it touches sensitive systems. The DeepSeek order is the AI piece of that strategy.
What it means for agencies and vendors
The practical work falls on IT teams. Agencies will have to find where DeepSeek is being used, including in tools they didn't realize relied on it, and remove it. Vendors that sell software to the government will need to check whether any part of their product is built on DeepSeek, because that now puts them at risk of losing federal business.
Expect agencies that were testing generative AI to lean toward providers that already have government-approved offerings. That's an analysis, not something the notice says, but it follows from how procurement tends to work: when one option is closed off, the rest benefit.
The ban also sets an example. Other bans, such as the one on Kaspersky software, were later echoed by state and local procurement rules, and some observers expect the same here.
What we still don't know
- The deadline. The reporting I found doesn't say how long agencies have to remove DeepSeek.
- DeepSeek's response. The Chinese Embassy in Washington hadn't commented when the news broke, and DeepSeek's own reaction wasn't in the coverage I reviewed.
- Evidence. DHS described risks, but hasn't shown public proof of data being taken.
- Next steps. Whether a broader measure, like a trade listing or a law passed by Congress, follows is unclear.
The bottom line
The federal government has decided it doesn't want a Chinese AI model running inside its own networks, directly or through its suppliers. That decision is firm and it has teeth, since it blocks both existing use and future purchases. But it's also limited. It covers government systems, it rests on risk assessments instead of proven incidents, and it leaves the wider US market alone for now.
The things to watch are the October 29 FCC vote, any move toward export controls, and whether DeepSeek or Beijing responds. Until then, the clearest takeaway is that the line has been drawn around federal networks, and it may not stay there.
