US Orders Federal Agencies to Remove DeepSeek AI Over Security Concerns

Washington has moved from caution to operational prohibition. On Tuesday, the Department of Homeland Security ordered federal civilian agencies to immediately stop purchasing products and services developed by DeepSeek and remove them from government IT systems. The Federal Register notice also covers third-party products and services built with DeepSeek technology, although agencies may request exceptions. That final provision is the policy’s most consequential detail: the government is not merely deleting an app or blocking a web domain; it is trying to remove an AI supply chain from sensitive systems.

src="https://images.unsplash.com/photo-1681164315990-b2a1e375eb69?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixid=M3w4OTkzMzh8MHwxfHJhbmRvbXx8fHx8fHx8fDE3OTI1MDY4NTN8&ixlib=rb-4.1.0&q=80&w=1080" alt="Cybersecurity professionals monitoring government computer systems"

The stated reason is data exposure. DHS warned that sensitive information submitted by US government users could potentially be accessible to Chinese authorities under Chinese law. That is a risk-based concern, not a finding that a particular breach occurred. Still, the distinction matters because agencies routinely feed contracts, case files, personnel information, cybersecurity details, and other sensitive material into AI tools. If operators or cloud partners cannot prove where data is stored, who can access it, how long it is retained, or whether it is used for training, the model becomes a governance risk regardless of output quality or price.

The mandate reaches downstream AI services

Covering downstream services is where the order acquires real enforcement teeth. DeepSeek’s open-weight models can be hosted by third parties, wrapped in APIs, embedded in productivity tools, or fine-tuned into industry-specific applications. Agencies therefore need more than a software inventory. They must trace model provenance, hosting locations, subprocessors, logging, telemetry, support access, and contractual data flows. Rebranding a DeepSeek-backed chatbot as an “AI assistant” would not satisfy the order. This also raises a difficult practical question: how can an agency verify what sits beneath a vendor’s claim that its service is independent?

A precedent built on earlier security restrictions

The action extends a trend rather than beginning one. A law passed in December 2025 targeted DeepSeek’s use in intelligence systems tied to national security; Tuesday’s order reaches the much larger civilian bureaucracy and covers both future purchases and installed tools. Beijing responded that Washington is overstretching national security and politicizing technology and trade, while vowing to protect Chinese companies’ rights. Separately, Chinese authorities are reportedly considering export controls on 23 AI-related technologies, including chip designs and advanced models. The result is a feedback loop: each side treats restrictions on access to chips and models as leverage, prompting the other to harden its digital borders.

For the market, US AI vendors gain a short-term substitution opportunity, while DeepSeek and any service provider dependent on it face compliance and reputational pressure. Federal buyers may receive favorable terms from alternative suppliers, yet rapid migration can also create rushed procurement, duplicated software, and new security blind spots. More important is the precedent. If DeepSeek can be excluded across civilian agencies, other governments may copy the model-by-model approach, pushing institutions toward locally hosted, auditable systems. Consumers are not directly barred from using DeepSeek; this order applies to federal civilian agencies. But the private market will likely feel the effects through cloud pricing, vendor due diligence, and stricter AI procurement clauses. The central takeaway is simple: in sensitive environments, where a model runs and who can reach its data now matter as much as how well it performs.