U.S. bans DeepSeek from federal agencies

The Department of Homeland Security issued a directive that all U.S. civilian federal agencies must remove DeepSeek AI software and any services that use its models within 60 days. The order bars the purchase of DeepSeek products and affiliated offerings, but does not apply to the Department of Defense or intelligence agencies. Agencies must withdraw DeepSeek from procurement catalogs, supplier inventories, and online acquisition portals. The directive is published in the Federal Register under docket DHS‑2026‑0628.

Reasons behind the ban

The order cites concerns over DeepSeek’s Chinese‑based servers, which could allow Chinese intelligence agencies to access data processed by the model. The directive also references the company’s account‑creation process, which collects personally identifiable information and tracks users through digital fingerprinting. The Department of Homeland Security warned that the model’s prompt‑response workflow might transfer sensitive information to servers in the People’s Republic of China, creating a potential intelligence‑gathering vector.

Agencies that believe a DeepSeek‑based tool is essential can submit a written exemption request through the Cybersecurity and Infrastructure Security Agency (CISA). The request must reach the relevant Homeland Security official within 60 days and explain why an exception is needed for government operations. Applicants must identify the products, give operational or national‑security grounds, and propose safeguards. While removal is under way, agencies are expected to strengthen network surveillance and investigate unusual activity, and CISA will help with technical implementation.

The directive also covers any third‑party tools or software built using DeepSeek implementations. DeepSeek has open‑sourced its inference framework, DeepInfer, which powers its DeepSeek‑V4.1‑Flash model, and some DeepSeek models are available on platforms such as Microsoft Azure and AWS. Analysts warn that enterprises may be embedding DeepSeek technology through vendors or platforms without knowing it, creating supply‑chain risks. They advise reviewing AI inventories, updating procurement requirements, and strengthening third‑party risk management.

DeepSeek continues to develop new models and draws enterprise interest for tasks such as document summarization and coding assistance. Its record includes a January 2025 leak of more than one million log entries, which the company reportedly addressed. Analysts say the open‑source framework could reduce costs, but its claims need validating, and risks such as jailbreaks and manipulation require careful assessment. The ban adds to scrutiny of AI supply chains, and analysts stress governance, restricted permissions, and monitoring when adopting new AI technologies.