The DeepSeek ban is now official: the Department of Homeland Security has ordered US civilian federal agencies to stop buying DeepSeek products and to remove any they already use. If you aren’t a federal employee or contractor, nothing changes for you today. But the order says a lot about how Washington now treats AI tools built abroad.
Here are the essentials, then what the order does and doesn’t cover.
- Who issued it: the Department of Homeland Security, under the Federal Acquisition Supply Chain Act.
- Who it hits: civilian federal agencies and anyone selling IT to them.
- What’s banned: products and services from DeepSeek’s parent company and its affiliates.
- Timing: the Federal Register notice publishes October 20, 2026.
- Who it skips: private companies and everyday users.
Those five lines are the whole story for most readers. The details below explain the “why” and the loose ends.
What the DeepSeek ban actually says
According to CyberScoop, agencies may not buy IT products or services from Hangzhou DeepSeek Artificial Intelligence or its affiliates. They must also pull any DeepSeek versions already in use. Engadget adds that DeepSeek is barred from taking part in federal contracts too.
The legal basis is the Federal Acquisition Supply Chain Act, which lets the government exclude vendors it sees as a supply-chain risk. CyberScoop reports that the ordering letter from Secretary Markwayne Mullin is dated June 26. So the policy has reportedly been in place for months, and the public notice is catching up.
No removal deadline has been reported. Agencies can ask for an exception, but they must file 60 days ahead with a “compelling justification,” such as a mission-critical task they can’t do any other way.
Why Homeland Security says DeepSeek is a risk
DHS points to two worries. The first is data: DeepSeek’s models may store information on servers in China, where authorities can compel access. The second is security, with the notice citing “multiple threat vectors” and the chance that models could “contain malware or other security vulnerabilities.”
These are DHS’s stated concerns, not findings published alongside the order. The notice doesn’t claim a specific breach of a federal system.
DHS also encourages, but doesn’t require, extra safeguards during removal. Think traffic monitoring, better logging, and flagging odd activity.
Who is affected, and who isn’t
The order is narrower than headlines suggest. This table sums up the reported scope.
| Group | Affected? | What it means |
|---|---|---|
| Civilian federal agencies | Yes | No new purchases; remove existing DeepSeek use |
| Federal contractors | Likely | DeepSeek can’t take part in federal contracts |
| Private companies | No | Not covered, per AI Weekly |
| Everyday users | No | The app and API remain legal to use |
Two gaps stand out. DeepSeek is the only Chinese AI model named, even though CyberScoop notes that Moonshot AI’s Kimi and Alibaba’s Qwen have drawn similar national-security concerns. And the order covers civilian agencies only.
The timing around DeepSeek’s latest release
The notice lands just days after DeepSeek shipped a new model. We covered the launch in our DeepSeek V4.1 Flash pricing breakdown, and the company’s cheap developer tools in our guide to the DeepSeek search API pricing.
The link between the two events is just timing. Engadget notes the order arrived a couple of days after the release, while CyberScoop’s reporting says the underlying letter dates back to June.
What the DeepSeek ban means for AI buyers
Agencies often buy AI through resellers and software suppliers, not straight from the model maker. That is why the wording about “third-party services built on” DeepSeek models matters, as AI Weekly describes it. A vendor that quietly runs DeepSeek under the hood could now be a problem for a federal buyer.
Expect more questions in procurement checklists: which model powers this tool, where does it run, and who can demand the data? Those were once niche questions. For government work, they are fast becoming standard.
Open-weight models add a twist. Because anyone can download and host them, a team might run a model on its own servers and avoid some data-location worries. Whether that satisfies the order is unclear from the reporting, so agencies will likely need guidance before relying on that route.
What you should do if you use DeepSeek
If you work for a federal agency or sell software to one, check with your compliance team before touching DeepSeek tools, including third-party apps built on its models. Treat this as a procurement rule, and expect contract language to change.
If you use DeepSeek on your own, the standard advice applies. Don’t paste passwords, client files, or personal data into any AI chatbot whose data policy you haven’t read. That holds for every provider, not just this one.
Frequently Asked Questions
Is the DeepSeek ban in effect for everyone in the US?
No. It applies to civilian federal agencies and their IT purchasing. Private companies and individuals can still use DeepSeek.
When does the DeepSeek ban start?
The Federal Register notice publishes October 20, 2026. CyberScoop reports the ordering letter is dated June 26, and no removal deadline has been reported.
Why did DHS ban DeepSeek?
DHS cites the risk that sensitive data could reach the Chinese government, and that the models might carry malware or other vulnerabilities.
Can agencies get an exception?
Yes, but they must apply 60 days in advance with a compelling justification, such as a mission-critical need with no alternative.
Our take
This ban is more about procurement than panic. For regular users it changes nothing, but it shows that where an AI model runs and who can compel its data is now a buying criterion. Expect more vendors, Chinese and otherwise, to face the same question, and expect businesses that sell to government to start asking it first.
Leave a Reply