The U.S. government is banning DeepSeek from the systems of its civilian federal agencies. The ban even extends to third-party tools that rely on the Chinese AI.
For months, DeepSeek has been at the center of a conflict that extends far beyond the question of who provides the most powerful or cheapest AI model.
Since the breakthrough of R1, the Chinese company has been in the crosshairs of U.S. authorities, regulatory agencies, and tech giants—in the midst of an increasingly fierce tech race.
Now Washington has tightened the screws even further: The U.S. Department of Homeland Security (DHS) is directing federal civilian agencies to remove DeepSeek’s technology from their IT systems.
What the U.S. government is specifically ordering
The scope of the measures initially affects a clearly defined area of the U.S. administration: the executive branch’s civilian agencies. The signed as early as July 24, 2026 and now in the Federal Register, this order compels these agencies to halt all new acquisitions of DeepSeek products and services.
Solutions already installed must be removed from the agencies’ IT systems. Exceptions or deadline extensions had to be requested in writing from DHS within 60 days of the order’s issuance; national security systems and strictly compartmentalized information systems of these agencies are exempt from this specific directive.
Third-party software is also in the crosshairs
DeepSeek is by no means limited to the simple web chatbot that you can access in your browser. The models can also be via application programming interfaces (APIs) to power features in software and cloud services developed by external teams.
In this way, the technology often works unnoticed in the background of tools that go by entirely different names. This is where the far-reaching scope of the decree comes into play: Even if a program is not labeled as DeepSeek, it falls under the ban as soon as it is hosted on a implementation of this technology.
The regulation uses this term but does not define it—for software providers, the exact scope therefore remains open to interpretation.
The Focus Is on Data Security
The greatest concern in Washington stems not only from the model’s capabilities but also relates to data flows. The Department of Homeland Security (DHS) emphasizes in the text of the regulation that the company operates its servers in China.
The services may collect personal data, create usage profiles, and analyze all content you enter in prompts.
The Department warns that, under Chinese law, this confidential information could be disclosed to local security authorities—with the risk that it could be used separately or combined with other sources to spy on U.S. citizens or government programs.
Not the First Move Against DeepSeek
U.S. leaders had already shown the AI from the Far East the door. A The law passed in December 2025 mandated the Director of National Intelligence to issue binding guidelines to ban DeepSeek applications and their successors from all national security systems—albeit with provisions for research purposes.
The key difference now lies in the scope: The new directive extends the ban comprehensively to all civilian federal agencies.
Serious Allegations of Unauthorized Distillation
The concerns go far beyond mere data protection issues. In September 2026 The NSA, FBI, and CISA sounded the alarm over large-scale distillation campaigns targeting cutting-edge U.S. AI models—and explicitly named DeepSeek.
Put simply, this method involves harvesting the responses from a pre-trained model to train one’s own system in a resource-efficient manner.
Technically, this approach is common practice, but the crux of the matter lies in unauthorized access: Both OpenAI as well as Anthropic accuse their Chinese competitor of targeting their systems with such data extraction.
Beijing Rejects the Allegations
In response, the leadership in Beijing accuses Washington of excessively inflating the concept of national security and of politicizing trade and technology issues. The Chinese Foreign Ministry announced that it would defend the legitimate rights and interests of domestic companies.
Beyond the diplomatic back-and-forth, this move underscores a fundamental shift: For government agencies, it is no longer merely a matter of which AI platforms they directly select, but rather which technology is powering the services and applications they purchase.
