The U.S. Department of Homeland Security issued an order on Monday prohibiting federal civilian agencies from using products developed by the Chinese artificial intelligence company DeepSeek, and requiring the removal of any existing installations. The directive, released as a PDF, mandates that agencies cease use of "all DeepSeek and DeepSeek-affiliate products, and third-party products and services developed using DeepSeek implementations." Compliance presents logistical challenges, as it remains unclear which third-party services were built using DeepSeek's technology or whether the phrase "developed using" applies to ongoing API access or merely to applications that incorporated DeepSeek tools during development.

The order also raises questions about AI gateway services such as OpenRouter and Vercel AI Gateway, which provide access to DeepSeek models through application programming interfaces. Agencies relying on these platforms may inadvertently consume DeepSeek models without explicit awareness. OpenRouter, for instance, "routes requests to the best available providers for your model," while Vercel AI Gateway selects default providers deemed optimal. Federal systems that have not restricted their AI service providers could therefore be utilizing DeepSeek models unintentionally.

U.S. officials have considered restricting DeepSeek since the release of the DeepSeek-R1 model in early 2025. On February 6, 2025, Representatives Darin LaHood (R-IL) and Josh Gottheimer (D-NJ) introduced legislation titled the No DeepSeek on Government Devices Act, citing national security concerns over data collection practices and potential ties to the Chinese government. The bill did not advance beyond committee, though several agencies had already implemented restrictions. The Department of Defense and related entities had blocked DeepSeek on military networks as early as January 2025, with additional actions taken by the U.S. Navy, NASA, and the House of Representatives. State-level prohibitions also preceded the DHS directive.

The timing of the DHS announcement may benefit domestic AI firms, including Anthropic, which are seeking public market opportunities. Open-source models from China have gained traction due to performance and cost efficiency, potentially limiting access for government users who rely on such tools. Meanwhile, security researchers have raised concerns about the mobile application associated with DeepSeek, noting issues such as unencrypted data transmission and connections to servers linked to Chinese technology firms. These vulnerabilities are not unique to DeepSeek but reflect broader challenges in mobile app security, particularly for applications integrating third-party advertising components.

Katie Moussouris, founder and CEO of Luta Security, stated that while national security considerations justify restricting DeepSeek's native app and API, the current order's scope exceeds practical enforcement capabilities. "Banning DeepSeek's own app and API is workable," she noted, "but the prohibition on 'third-party products and services developed using DeepSeek implementations' lacks definitional clarity. The term could encompass any software derived from DeepSeek or even applications that used the company's tools during development. No standardized AI bill of materials exists to track such dependencies, making verification unfeasible for most vendors."

Moussouris further criticized the order's expansive reach, arguing that restrictions applied to U.S.-hosted versions of DeepSeek models address few of the cited risks. "The concerns primarily involve data transmission to servers in China and app-based data collection," she explained. "Hosting the model domestically mitigates most of these issues. Regarding the model itself, the order cites potential malware or vulnerabilities, a characteristic shared by all software. NIST assessments indicate DeepSeek exhibits higher susceptibility to prompt injection and jailbreaking than domestic counterparts, though all models face similar challenges in this domain."

The directive underscores growing scrutiny over foreign-developed artificial intelligence systems within U.S. government infrastructure, reflecting broader tensions between technological accessibility, cybersecurity imperatives, and geopolitical considerations.