US DHS Bans DeepSeek AI Across Federal Civilian Agencies
New directive mandates removal of all DeepSeek-affiliated AI products from federal government systems.
1DHS Issues Sweeping Ban
The US Department of Homeland Security has officially ordered all federal civilian agencies to remove DeepSeek technology products. This directive covers direct implementations and any third-party services developed using the Chinese lab's AI. The order aims to mitigate potential national security risks associated with data capture and foreign government access.
2Ambiguity in Compliance Scope
The directive creates significant operational uncertainty regarding third-party software. It remains unclear if the ban applies to apps coded with DeepSeek assistance or only those using its models via API. Agencies now face the difficult task of auditing complex software supply chains to ensure no DeepSeek-derived components remain in their systems.
3Gateway Service Risks
AI gateway services like OpenRouter and Vercel AI Gateway pose a hidden compliance challenge for agencies. These platforms automatically route requests to the best available model, potentially invoking DeepSeek without explicit user intent. Agencies must now implement strict provider constraints to prevent inadvertent usage of prohibited models within their internal workflows.
4Legislative and Defense Precedent
This civilian ban follows earlier actions by the US Defense community, including the Pentagon’s IT networks and NASA. Lawmakers Darin LaHood and Josh Gottheimer previously introduced the No DeepSeek on Government Devices Act in February 2025. These collective efforts reflect a broader, multi-agency strategy to isolate Chinese AI from government infrastructure.
5Market Impact on Competitors
The exclusion of cost-effective, high-capability Chinese models may benefit American AI firms like Anthropic. By restricting access to open-weight alternatives, the government effectively narrows the vendor landscape. This shift could force agencies to rely on domestic model makers, potentially increasing procurement costs for taxpayers while limiting the diversity of available AI tools.
6Mobile App Security Concerns
Reports from December 2025 highlighted critical vulnerabilities in DeepSeek’s mobile applications, including plaintext data leaks and insecure server connections. While these issues mirror common problems in third-party advertising SDKs, the government views them as unique threats. The primary concern remains the ingestion of sensitive government prompts and local secrets by AI agents.
7Practicality of Enforcement Challenges
Security expert Katie Moussouris argues that the order’s breadth exceeds practical enforcement capabilities. Requiring vendors to attest that their code never touched DeepSeek during development is currently impossible without an AI software bill of materials. This creates a compliance burden that may be technically unverifiable for most federal IT departments today.
8Vulnerability and Risk Assessment
NIST findings indicate that DeepSeek models are susceptible to prompt injection and jailbreaking, though these are industry-wide challenges. The order prioritizes preventing data flow to servers in China over addressing model-specific vulnerabilities. Ultimately, the government is prioritizing risk avoidance over the potential utility of these models in sensitive federal operations.