大模型网摘deepseek.wang

News · 2026-10-22

US DHS Issues Exclusion Order: Federal Civilian Agencies Told to Drop DeepSeek Products

On October 20, 2026, the US Department of Homeland Security published Federal Register document 2026-21447: Exclusion and Removal Order #2026-01, instructing federal civilian executive branch agencies to stop procuring DeepSeek products and to remove them from their IT systems, with both requirements effective immediately.

Key terms of the order

According to the AI News Report's transcription of the signed order, it was signed by DHS Secretary Markwayne Mullin and copied to OMB Director Russell Vought. It applies to federal civilian executive branch agencies — national security and compartmented systems are excluded, with military and intelligence systems under other authorities. Notably, the order reaches beyond DeepSeek's own apps: third-party products and services developed using DeepSeek are also covered. The signed order carries an original release date of July 24, 2026; it entered public inspection on October 19 and was published on October 20. The legal basis is the Federal Acquisition Supply Chain Security Act of 2018.

Exceptions and gray zones

Agencies that cannot comply can request an exception in writing within 60 days, reviewed by CISA. The order's text does not name open weights, self-hosted copies, or AI gateways — precisely the places where a DeepSeek model can sit inside a stack unnoticed, which is where an inventory should start, as the explainer notes. TechSpot reported that the order prompted an angry response from China.

Practical impact

This is a US federal procurement restriction: it directly governs federal civilian agencies, not DeepSeek's public services elsewhere. Two groups should assess exposure: SaaS vendors selling into the US market with DeepSeek models embedded in their products, and companies that must document their model supply chain for US customers. Self-hosting open weights is one path the order leaves unnamed — see our analysis of self-hosting scenarios with DeepInfer. For the model's recent launch, see the V4.1-Flash-1017 release notes, and for substitution options, the October alternatives roundup.