The Department of Homeland Security has told federal civilian agencies to stop buying from DeepSeek and to pull the Chinese AI lab’s technology out of government computer systems, TechCrunch reported on 20 October. The order is narrower than a ban on the company and wider than a ban on its apps. For anyone who sells software to Washington, that gap is the whole story.
The scope is federal civilian agencies. Private citizens and businesses are not covered, and this is not an import or export control. Agencies can request exceptions, though TechCrunch’s report does not say how those requests are handled or who decides them.
The widest clause reaches outside vendors. Products and services from third parties that were built using DeepSeek technology fall under the order too. Picture a contractor whose support chatbot or document tool runs on a DeepSeek model underneath: the DeepSeek name may appear nowhere on the invoice, yet the product still has a procurement problem. Because DeepSeek competes with OpenAI and Meta on low-cost open-weight models, which any developer can download and build on, that kind of hidden dependency is easy to create and easy to forget. Vendors now need an accurate inventory of the models inside their own stack.
DHS gave a data-handling reason. It raised concerns about how DeepSeek treats user data. Its warning rests on Chinese law: under it, Chinese authorities “could potentially” gain access to material from US government users that is sensitive. That hedge is DHS’s own. It describes a risk, and nothing in the reporting alleges that government data was actually accessed.
Beijing objected. At an official briefing on Tuesday, Foreign Ministry spokesperson Mao Ning objected to “overstretching the concept of national security and politicizing technology and trade issues.” Mao added: “We will resolutely safeguard the legitimate rights and interests of Chinese enterprises.”
The order extends earlier limits. A law passed in December 2025 went after DeepSeek’s use inside intelligence agencies’ national security systems. The new order moves the restriction to civilian agencies and covers both new purchases and systems already running.
A separate set of accusations surrounds the company, and each has an accuser. A September report said the US accused DeepSeek and five other Chinese AI companies, Alibaba and Moonshot AI among them, of distillation at scale, which means training one model on another model’s answers, to pull information out of American systems. Earlier this year OpenAI informed US lawmakers that it had seen what it suspected were DeepSeek attempts to distill its models. Anthropic, for its part, said it had spotted unauthorized efforts by DeepSeek and other Chinese labs to harvest Claude’s outputs. These are allegations, and two of the three come from companies that compete with DeepSeek directly. They also differ from DHS’s stated reason, which concerns data handling, not distillation.
The backdrop is a widening tech contest. Reportedly, China is considering curbs on exports of 23 technologies related to AI, Bloomberg said last week. Certain advanced models and chip designs are among them. Governments around the world have scrutinized DeepSeek since January 2025, when its R1 reasoning model rattled the industry and sent Nvidia shares into a sharp sell-off.
Procurement rules tend to travel past the agencies that write them, because suppliers prefer one clean stack to two. Reporting on the order mentions no deadline, no penalty, and no list of affected vendors, so any company with federal civilian customers should audit which models sit inside its products now, before a contracting officer asks.
Reported by TechCrunch (Kate Park), 20 October 2026.
