DHS Orders Federal Agencies to Strip DeepSeek From Their Systems

The Monday order formalizes a July directive and marks the broadest federal lockout yet of the Chinese AI startup's technology.

The Trump administration has told civilian federal agencies to rip DeepSeek out of their networks entirely, making public on Monday a Department of Homeland Security order that bans the Chinese AI company's technology from government IT systems.

The directive requires federal civilian agencies to remove any DeepSeek-related technology already running on their systems and prohibits them from buying DeepSeek products or services going forward. That prohibition isn't limited to DeepSeek's own apps and models. It also covers third-party products built using the company's technology, closing off a potential workaround for agencies that might otherwise access DeepSeek's AI through an intermediary vendor.

The order itself isn't brand new. DHS originally issued it on July 24; Monday's move made it public. It applies broadly across the federal civilian government, naming agencies including the Treasury, State and Justice departments. One notable carve-out: the Department of Defense is excluded, meaning the ban is specifically a civilian-agency measure rather than a government-wide one.

DHS laid out its reasoning in blunt terms, stating that DeepSeek's AI models "present multiple threat vectors." The agency's core worry is data exposure. It warned that running DeepSeek's models could let sensitive U.S. government data flow back to China, potentially handing Chinese authorities intelligence on American citizens and federal programs. DHS also flagged technical vulnerabilities, pointing to open-source research that found multiple critical security flaws in DeepSeek's iOS app.

Agencies aren't without recourse. Executive agencies can seek exceptions to the order, including compliance delays, by filing a written justification with DHS within 60 days of the order's issuance.

A year of escalating restrictions

Monday's order is the latest and most sweeping entry in a pattern that's been building since DeepSeek first grabbed global attention at the start of 2025. The startup's rapid rise rattled U.S. tech and policy circles, and government agencies have been peeling back access to its tools in stages ever since.

The Defense Information Systems Agency moved first, blocking access to DeepSeek models in January 2025, according to Bloomberg reporting cited in the CNBC article. Commerce Department bureaus followed in March, telling staff that DeepSeek was off-limits on government devices, per Reuters. By December 2025, Congress had restricted DeepSeek's use within intelligence community systems altogether.

Each of those actions targeted a specific slice of government. DHS's order is broader by design, functioning as a civilian-agency-wide mandate rather than a department-by-department patchwork.

AI as a front line in the U.S.-China rivalry

The timing reflects where things stand between Washington and Beijing right now. Artificial intelligence has become one of the sharpest points of friction in the broader technological rivalry between the two countries, with both governments treating leadership in AI as a matter of national security as much as commercial advantage.

Framed that way, DHS's concerns aren't just about app security. The underlying anxiety is that a Chinese-developed AI system embedded in U.S. government workflows could become a channel for data collection, whether through deliberate design or exploitable vulnerabilities. That's consistent with the broader U.S. posture toward Chinese tech companies in sensitive sectors over the past several years, extended now to generative AI.

DeepSeek and China's Embassy in the U.K. did not respond to CNBC's request for comment.

What to watch

The 60-day exception window gives agencies some room to maneuver, so the real test will be how many request delays or carve-outs and on what grounds. It's also worth watching whether this civilian-focused order becomes a template for further action, or whether the Pentagon's exclusion signals a more fragmented approach to how different parts of the U.S. government handle Chinese AI tools going forward.