US Bans DeepSeek From Civilian Federal Agencies, Citing Data Security and Cybersecurity Risks

The Trump administration has barred Chinese artificial intelligence startup DeepSeek from civilian US government agencies, escalating Washington's efforts to restrict Chinese technology in sensitive government systems as competition between the two countries increasingly centers on AI capabilities, data security and technological independence.

The US Department of Homeland Security made public an order on Monday requiring federal civilian agencies to remove DeepSeek-related technology from their information technology systems and prohibiting them from procuring the company's products and services. The restrictions also extend to third-party products and services developed using DeepSeek technology.

Originally issued on July 24, the order applies to civilian federal agencies, including the departments of the Treasury, State, and Justice, but excludes the Department of Defense.

The administration's decision reflects concerns that DeepSeek's AI models could expose sensitive government information to unauthorized access, create cybersecurity vulnerabilities, or provide Chinese authorities with information about US citizens and government programmes.

The order adds to a series of restrictions imposed by American government bodies since DeepSeek attracted international attention in early 2025. It also highlights a growing distinction in Washington's approach to AI: the technology is being assessed not only for its capabilities and cost, but also for the national security risks associated with its ownership, development and deployment.

The Department of Homeland Security said DeepSeek's models "present multiple threat vectors," citing concerns about potential access to sensitive information by Chinese authorities and vulnerabilities that could compromise government systems.

According to the order, using DeepSeek models could result in sensitive US government data being transferred to China. Such information could potentially reveal details about American citizens, government operations, and public programmes.

The agency also warned that the models could contain malware or other security weaknesses. It cited open-source research identifying multiple critical vulnerabilities in DeepSeek's iOS application as part of its justification for the restrictions.

These concerns place data handling and software security at the center of the government's decision. AI systems can process sensitive information through prompts, uploaded documents, application integrations, and other interactions. Where models are accessed through external services, the way data is collected, stored, and processed becomes an important consideration for organizations handling confidential material.

The potential risks extend beyond the underlying AI model. Third-party applications incorporating a model can introduce additional software dependencies, data flows, and security weaknesses. By including products and services developed using DeepSeek technology, the order seeks to prevent agencies from continuing to use the company's capabilities indirectly through other vendors.

However, the stated concerns should be distinguished from publicly established evidence of a specific breach of US government systems. The order identifies potential security threats and cites vulnerabilities, but the information provided does not establish that DeepSeek transferred classified government information to Chinese authorities or that its technology was responsible for a confirmed intrusion into federal systems.

The restrictions instead reflect the administration's assessment of the risks associated with allowing the technology into government IT environments.

Executive agencies can request exceptions, including additional time to comply, by submitting written justifications to the Department of Homeland Security within 60 days of the order's issuance. That provision gives the government a mechanism to consider circumstances in which agencies believe continued use or a delayed transition is necessary.

Restrictions Build on Earlier Federal Action

The latest order follows several measures introduced by US government agencies since DeepSeek's emergence in early 2025.

In March 2025, bureaus within the US Commerce Department told staff that DeepSeek was prohibited on government devices. The Defense Information Systems Agency also moved to block access to DeepSeek models in January 2025, according to reports from Bloomberg.

Congress subsequently restricted the use of DeepSeek in intelligence community systems in December 2025.

The measures have developed across different parts of the federal government, with the latest Department of Homeland Security order extending restrictions across civilian agencies and addressing indirect use through third-party products and services.

The latest order explicitly excludes the Department of Defense, even though separate restrictions affecting defense-related systems had already been reported. This means the new directive should not be interpreted as the first US government action against DeepSeek or as a single measure covering every federal system.

The broader policy direction is nevertheless clear: Washington is seeking to reduce the presence of technology it considers potentially vulnerable to foreign government access within systems that handle sensitive information.

Against that backdrop, there will be consequences beyond the direct purchase of AI models for government technology procurement. Agencies may need to review software vendors, application integrations, and underlying technology dependencies to establish whether prohibited capabilities remain embedded in their systems.

Such reviews can increase compliance costs and complicate procurement, particularly where AI tools have been integrated into broader software products rather than installed as standalone applications.

Beijing Rejects The Politicization of Technology

China has pushed back against US restrictions on its technology companies, framing such measures as an inappropriate extension of national security concerns into commercial and scientific activity.

Asked by Reuters about the latest move, China's foreign ministry said Beijing would safeguard the rights and interests of its companies.

Ministry spokesperson Mao Ning said that "China consistently opposes the over-generalization of national security as well as the politicization of science, technology and trade issues," according to Reuters.

The response reflects the wider dispute between Washington and Beijing over access to markets, technology supply chains and the role of national security in commercial competition.

For the United States, restrictions on foreign AI systems are presented as safeguards against potential data exposure and cyber threats. China, meanwhile, objects to measures it views as politicizing technology and limiting the commercial activities of its companies.

The dispute is unfolding as AI becomes an increasingly important component of economic competitiveness and national security. Governments are evaluating not only the performance of AI systems but also the risks associated with where they are developed, how they handle data, and whether their software can be trusted within sensitive environments.

DeepSeek's emergence at the beginning of 2025 intensified that debate by drawing attention to China's growing AI capabilities. Its subsequent exclusion from government systems underpins how quickly concerns about the strategic implications of AI can translate into procurement restrictions.

The consequences could extend to companies supplying government agencies, particularly vendors whose products incorporate models or other technology components developed by third parties. They may face greater scrutiny of their AI supply chains and pressure to demonstrate where data is processed, what external systems are involved, and how security vulnerabilities are managed.

The order limits access to a significant category of DeepSeek's potential customers in the United States and could complicate efforts by other vendors to offer its technology within federal civilian systems.