The U.S. Department of Homeland Security (DHS) issued a directive on Monday ordering civilian government agencies to remove DeepSeek products and any third-party services built on the model from their IT systems, while also prohibiting new purchases. The notice cites security risks, stating that using DeepSeek "may result in the transfer of sensitive data to the PRC" and could generate "actionable intelligence on U.S. persons and/or government agencies." This order expands upon a July memo that had already instructed agencies to monitor network traffic for DeepSeek-related risks.
The ban applies broadly, reaching beyond DeepSeek’s own application to include external services developed using its open-source architecture. While DeepSeek is an open-source model capable of running locally or via non-Chinese providers, DHS highlights concerns regarding data handling by the Chinese developers who host versions of the service on servers located in China. Prior to this federal directive, several entities had already restricted DeepSeek, including the Commerce Department, the Navy, and state governments in New York, Tennessee, and Virginia.
DeepSeek, a Hangzhou-based startup, gained significant attention in January 2025 when its low-cost, high-performance model caused a sharp decline in Nvidia’s stock value. The company has continued to release updates, including the V4-Pro preview in April, which is priced significantly lower than competitors like OpenAI’s GPT-5.5 Pro. The DHS action coincides with broader regulatory scrutiny of Chinese technology; notably, the Federal Communications Commission is scheduled to vote on October 29 regarding restrictions on Chinese labs testing electronics for U.S. use.
This directive marks a decisive escalation in the U.S. government's effort to decouple critical infrastructure from Chinese AI models, moving from passive monitoring to active exclusion. By targeting not just the core product but also third-party integrations, DHS acknowledges the pervasive nature of open-source AI supply chains. The emphasis on data sovereignty suggests that the primary concern is not merely technical performance but the potential for state-level intelligence gathering through user interactions and backend server architectures hosted in China.
For the broader market, this action reinforces the bifurcation of global AI development into distinct geopolitical spheres. While DeepSeek continues to offer cost-effective solutions that challenge Western incumbents, its utility within U.S. public sector workflows is now effectively nullified. This may accelerate the adoption of domestically sourced or allied-nation AI alternatives among government contractors and vendors who previously considered integrating DeepSeek capabilities. The upcoming FCC vote further signals a coordinated inter-agency strategy to limit Chinese technological influence across both software and hardware domains.
