Trump Administration Bans DeepSeek AI From Federal Civilian Agencies

The Department of Homeland Security has instructed civilian federal agencies to remove the Chinese artificial intelligence system DeepSeek from their computers and halt any further purchases of its products. The order, signed July 24 by DHS Secretary Markwayne Mullin and released publicly this week, applies to software developed by DeepSeek, its affiliates, and any third-party vendors that incorporate the technology. Agencies may request limited exceptions, but final approval rests with DHS.

DeepSeek, headquartered in Hangzhou with servers located in China, gained attention starting in early 2025 for its open-source models. The ban focuses on the risk that Chinese law grants authorities broad access to data held by private companies. The DHS directive states that DeepSeek models present multiple threat vectors because servers sit inside the PRC, where the legal system permits extensive government cyber and intelligence access to private infrastructure. User prompts processed by the system can feed into data collection that includes personnel records, internal communications, and agency planning documents.

The restriction covers civilian agencies such as the Treasury, State, and Justice departments but does not extend to the Department of Defense. It builds on earlier limits, including a December 2025 congressional measure that barred DeepSeek from intelligence community systems. The move occurs against a backdrop of intensifying U.S.-China competition over artificial intelligence, where control of data flows and model infrastructure has become a central policy concern.

Chinese officials responded by rejecting the characterization of the ban. A foreign ministry spokesperson told Reuters that Beijing would protect the rights of its companies and opposed treating national security concerns as a blanket justification for restricting technology and trade. The statement framed the U.S. action as an example of politicizing technical issues rather than addressing specific, evidence-based vulnerabilities.

For agencies that have already integrated DeepSeek tools into workflows, the order requires a phased removal process. Officials must identify any third-party services that rely on the Chinese models and replace them with alternatives cleared by DHS. The policy reflects a wider pattern in which governments treat AI systems hosted in authoritarian jurisdictions as potential vectors for espionage, especially when those systems handle prompts that could reveal operational details.

Analysts note that the decision aligns with prior steps taken by both Republican and Democratic administrations to limit exposure of sensitive networks to Chinese technology. The emphasis on civilian agencies underscores the volume of non-classified but still sensitive information that flows through routine government operations. Personnel data, procurement plans, and policy drafts, once entered into an AI interface connected to servers in China, become subject to collection under PRC national intelligence laws.

The ban leaves open the possibility that agencies could continue limited testing or evaluation under strict controls, provided DHS grants an exception. It also signals that future procurement decisions will face heightened scrutiny when vendors incorporate foreign AI components. As federal departments adjust their technology stacks, the episode illustrates how quickly new AI tools can move from experimental use to security liabilities once their data-handling practices come under review.