The US government just widened its DeepSeek problem from a procurement footnote into a cleanup job. On October 20, the Department of Homeland Security published an order telling federal civilian agencies to stop buying products and services from the Chinese AI startup and to remove what they already run. The part that will keep IT teams busy is the last clause: it also covers third-party products and services built using DeepSeek technology.
The short version
- DHS ordered civilian agencies to exclude DeepSeek from procurement and remove it from existing IT
- The order reaches affiliated entities and third-party products built with DeepSeek implementations
- Agencies can request exceptions
- The stated reason is national security and data handling, not a specific breach
- Beijing says the US is “overstretching” national security
What the order actually says
The notice appeared in the Federal Register on October 20 under docket DHS-2026-0628. It was issued under the Federal Acquisition Supply Chain Security Act of 2018, known as FASCSA, the law that lets the government exclude and remove technology it judges to be a supply chain risk. According to the notice, the Secretary of Homeland Security acted after reviewing a recommendation from the Federal Acquisition Security Council and the evidence attached to it.
Two things happen at once. First, IT products and services developed by Hangzhou DeepSeek Artificial Intelligence Co., Ltd, including affiliated entities, are excluded from federal civilian procurement. Second, everything already in place has to come out: DeepSeek and affiliate products, and “third-party products and services developed using DeepSeek implementations.”
The stated concern, as TechCrunch reported, is data. DHS warned that sensitive information from US government users could potentially be accessible to Chinese authorities under Chinese law.
Why the third-party clause matters most
A ban on one vendor’s app is easy to follow. A ban on anything built with that vendor’s technology is not. DeepSeek publishes open-weight models, which means other companies can download them, fine-tune them and wrap them in their own products. A document summarizer, a coding assistant or a customer-support bot sold to an agency might run a DeepSeek-derived model without saying so on the box.
The practical headache
Agencies will need to ask their software vendors what models sit underneath their AI features. Many vendors have never been asked that question, and some may not have a clean answer ready.
How the restrictions stack up
| Step | What it covered |
|---|---|
| December 2025 law | DeepSeek use in intelligence national security systems |
| September 2026 federal report | Accused DeepSeek and five other Chinese AI firms of large-scale model distillation |
| October 20, 2026 DHS order | Civilian agencies: no new purchases, removal from existing systems, third-party products included |
The distillation backdrop
The order lands after a September federal report accused DeepSeek and five other Chinese AI companies, including Alibaba and Moonshot AI, of using large-scale distillation to pull capability out of American models. Distillation means training one model on the outputs of another. We broke down that advisory when the NSA named six Chinese AI companies it says were draining Claude, GPT and Gemini. OpenAI and Anthropic have both said separately that they saw suspected DeepSeek attempts to collect model outputs.
That context helps explain the timing, though the DHS order itself rests on supply chain and data security grounds rather than on the distillation claims.
Beijing pushes back
China’s Foreign Ministry spokesperson Mao Ning said on Tuesday that Beijing opposes “overstretching the concept of national security and politicizing technology and trade issues,” and promised to “resolutely safeguard the legitimate rights and interests of Chinese enterprises.” Bloomberg also reported last week that China is considering export restrictions on 23 AI-related technologies, including chip designs and certain advanced models. Whether the two stories are linked is not established, but both point the same way: AI is now handled as a trade and security asset, not just a product category.
The uncomfortable part: the model is good
None of this changes how competitive DeepSeek has become. Our earlier look at the benchmarks showed that DeepSeek is now about 3% behind the best US model, down from a 15% gap eight months earlier. Cheap, strong open-weight models are exactly why agencies, and the vendors that sell to them, were tempted to use it in the first place. The order removes that option for civilian federal systems. It does not touch what private companies or individuals run on their own.
What to watch next
- Exceptions. Agencies can request them, so the first approved cases will show how strict the order really is.
- Vendor disclosures. Expect software suppliers to start stating which models power their AI features.
- Other Chinese labs. The same FASCSA tool could be used against other vendors named in the September report.
- Retaliation. China’s export control discussions could raise the cost of the ongoing tech split.
The bottom line
The DeepSeek order is less about one chatbot and more about a principle: if a model comes from a vendor the government distrusts, so does everything built on top of it. That is a hard rule to enforce in an industry where models are routinely rebranded, fine-tuned and bundled. Federal IT teams now have to find out what they actually run, and vendors have to be ready to tell them.
Sources and further reading
- TechCrunch: US orders federal agencies to remove China’s DeepSeek AI from their systems
- Federal Register: Notice of Order Regarding Exclusion and Removal of DeepSeek Products
About this article: GeekBlog covers U.S. technology news, AI, phones, smartwatches and gaming. Every story is written and checked under our Editorial Policy. Spotted a mistake or have a story tip? Contact our editors.
