DHS Bans Chinese AI Model DeepSeek from Federal Civilian Agencies

The Department of Homeland Security (DHS) has issued a ban on the use of the Chinese artificial intelligence model DeepSeek and its associated products by civilian federal agencies. This directive, slated for publication in the Federal Register, invokes the Federal Acquisition Supply Chain Act and recommendations from the Federal Acquisition Security Council to prohibit federal agencies from procuring any information technology products or services from Hangzhou DeepSeek Artificial Intelligence, the parent company.

Secretary of Homeland Security Markwayne Mullin detailed the rationale behind the ban in a letter dated June 26, revealing that the exclusion has been in effect for several months. Mullin asserted that DeepSeek's AI models pose "multiple threat vectors" to the IT systems of civilian federal agencies. These threats are partly attributed to the company's origin, raising concerns about potential privacy violations and mass data collection due to data storage practices in China, which is subject to the country's national security laws.

According to Mullin, DeepSeek models store user data on servers located in China. This location subjects the data to Chinese national security laws, which grant the government broad authority to access information held by domestic companies. The models also collect personally identifiable information from users and track their "digital fingerprints" online. Furthermore, the standard "prompt and response" process inherent in AI models like DeepSeek could lead to the transfer of sensitive data to the Chinese government.

Mullin emphasized the potential national security implications, stating that the transfer of such information to the People's Republic of China, whether alone or combined with publicly available data, could yield actionable intelligence on U.S. persons or government programs. This action comes over a year after U.S. AI companies and federal officials accused DeepSeek's creators of "systematically" distilling their models.

The ban mandates that civilian agencies cease all future procurements of DeepSeek products and remove any existing installations. While federal agencies have historically exercised discretion in blocking certain AI models, this formal directive establishes a clear prohibition. Agencies may seek exceptions, but these require a 60-day advance request with a "compelling justification," such as an inability to fulfill mission-critical functions or other national security imperatives.

In addition to removal, agencies are "encouraged" to bolster their security controls to mitigate any potential compromises stemming from the use of DeepSeek. Mullin advised agencies to monitor network traffic, enhance logging and data exfiltration controls, and detect anomalous activity on relevant systems during the transition period.

While DeepSeek is the only Chinese AI model specifically named in this order, it highlights broader concerns among national security officials regarding several Chinese open-weight AI models, including those from Moonshot AI (Kimi) and Alibaba (Qwen), which have faced accusations of illegally distilling U.S. models. This DHS ban underscores the growing scrutiny of foreign AI technologies and their potential impact on U.S. federal systems.