DeepSeek deep-sixed at all US federal civilian agencies

US federal civilian agencies must now stop using DeepSeek AI products, potentially incurring costs for replacement and integration. The broad nature of the ban also creates compliance challenges for agencies using third-party services that might inadvertently incorporate DeepSeek technology.

The U.S. Department of Homeland Security (DHS) has issued a directive prohibiting federal civilian agencies from using any technology products from Chinese AI lab DeepSeek. The order, effective immediately, also mandates the removal of any existing DeepSeek implementations. This ban extends to DeepSeek-affiliated products and any third-party products or services developed using DeepSeek implementations. The directive's broad language raises questions about what constitutes "developed using" and whether it includes services that route requests through DeepSeek models via APIs, such as OpenRouter and Vercel AI Gateway. This action follows earlier bans by the Defense Information Systems Agency (DISA), the U.S. Navy, NASA, and the U.S. House of Representatives, driven by national security concerns that DeepSeek's AI models could collect data on U.S. citizens and transmit it to the Chinese government. The DHS order, while taking time to reach civilian agencies, was anticipated following concerns raised as early as 2025.

The directive's vagueness regarding "third-party products and services developed using DeepSeek implementations" presents a significant compliance hurdle for federal agencies. It remains unclear if this encompasses software that merely used DeepSeek for coding assistance or models derived from DeepSeek, creating a potential compliance audit nightmare. This ambiguity could lead to agencies over-blocking services or struggling to identify all affected technologies, potentially disrupting critical operations.

This ban, while addressing national security concerns, could inadvertently benefit domestic AI companies by reducing competition from potentially lower-cost Chinese AI models. Agencies that cannot deploy DeepSeek or similar offerings may be compelled to use more expensive U.S.-based alternatives, impacting taxpayer budgets. Furthermore, the order's broad application to US-hosted versions of DeepSeek models, even if they mitigate data exfiltration risks, highlights a risk-averse approach that prioritizes control over nuanced security assessments.