US Federal Agencies Ordered to Remove DeepSeek AI Technology
Homeland Security mandates removal of Chinese AI tools citing national security and data privacy risks.
1DHS Issues Broad Ban
The US Department of Homeland Security has officially directed all federal civilian agencies to remove DeepSeek technology products. This mandate extends beyond direct software, covering any third-party services developed using DeepSeek implementations. The order aims to mitigate potential national security threats associated with data capture and transmission to foreign-based AI labs.
2Ambiguity in Compliance Scope
The directive creates significant operational uncertainty regarding the definition of products developed using DeepSeek. It remains unclear if the ban applies to ongoing API access or simple code assistance. Agencies now face the difficult task of auditing software supply chains to ensure no indirect reliance on DeepSeek models exists within their systems.
3Risks of AI Gateways
Services like OpenRouter and Vercel AI Gateway pose a hidden compliance risk for federal agencies. These platforms automatically route requests to various providers, potentially utilizing DeepSeek models without explicit user intent. Unless agencies actively constrain their AI gateway configurations, they risk inadvertent usage of prohibited models, complicating their adherence to the order.
4Legislative and Defense Precedent
This order follows earlier actions by the US Defense community, including the Pentagon, Navy, and NASA, which began blocking DeepSeek in January 2025. Legislative efforts, such as the No DeepSeek on Government Devices Act introduced by Representatives LaHood and Gottheimer, previously highlighted concerns that these AI models capture sensitive data on US citizens.
5Market Impact and Competition
The ban may inadvertently benefit American AI companies by removing a low-cost, high-capability competitor from the government market. Agencies restricted from using open-weight models from China will likely shift toward domestic providers. This transition could lead to higher procurement costs for taxpayers as agencies rely exclusively on favored US-based model makers.
6Mobile App Security Concerns
Reports from December 2025 identified critical vulnerabilities in DeepSeek’s mobile applications, including plaintext information leaks and connections to Chinese servers. While these issues mirror common mobile security flaws found in many third-party SDKs, the government is prioritizing the specific risks of prompt ingestion and sensitive data handling by AI agents.
7Practicality of Enforcement Challenges
Luta Security CEO Katie Moussouris argues that the order’s breadth makes verification nearly impossible for vendors. Since current software bills of materials do not track developer tools, vendors cannot easily attest that their code never touched DeepSeek. This creates a compliance burden that may exceed the actual security benefits of the policy.
8Technical Vulnerability Assessment
NIST evaluations suggest DeepSeek models are more susceptible to prompt injection and jailbreaking compared to US alternatives. While the order cites these vulnerabilities, critics note that all AI models share similar risks. The government’s focus remains on data flow to China, regardless of whether the models are hosted on US servers.