US Orders Federal Agencies to Remove China’s DeepSeek AI From Their Systems

The United States has ordered federal civilian agencies to remove products developed by Chinese artificial intelligence company DeepSeek from their information technology systems and stop purchasing its products and services. The directive, issued by the Department of Homeland Security (DHS) and made public on October 19–20, 2026, also covers third-party products and services developed using DeepSeek implementations. The government says the restrictions address national security risks, particularly the possibility that sensitive information handled by government employees or contractors could become accessible to Chinese authorities.

The decision expands restrictions on DeepSeek within the US government as Washington and Beijing compete for leadership in artificial intelligence. Unlike earlier measures that focused on particular government devices or national security systems, the new order applies across federal civilian agency procurement and IT systems. It does not create a blanket prohibition on DeepSeek for private individuals or businesses. China has condemned the move, arguing that the United States is using national security concerns to restrict Chinese technology companies.

Key Takeaways

• Federal restrictions: US civilian agencies must remove DeepSeek products from their IT systems and stop purchasing them.
• Third-party coverage: The order also covers products and services developed using DeepSeek implementations.
• Security rationale: DHS warns that sensitive government data could potentially be exposed to Chinese authorities.
• Legal basis: The order invokes the Federal Acquisition Supply Chain Security Act of 2018.
• Scope: The directive applies to federal civilian agencies, not a nationwide ban on private use.
• China’s response: Beijing has criticised the restrictions and called for them to be revoked.
• Broader significance: The decision adds to the technology and security tensions between the United States and China.

What Has the US Government Ordered?

The DHS directive requires federal civilian agencies to exclude DeepSeek products and services from future procurement and remove covered products from their information technology environments.

The order extends beyond DeepSeek’s own chatbot and applications. It also includes affiliated entities and third-party products and services developed using DeepSeek implementations.

That broader language is significant because government agencies increasingly obtain software through contractors, cloud platforms and technology vendors. A service may incorporate an AI model without presenting itself as a standalone chatbot, making it more difficult for an organisation to identify every place where the technology is used.

The order was published in the Federal Register under the title Notice of Order Regarding Exclusion and Removal of DeepSeek Products From Federal Civilian Executive Branch Information Technology Systems and Procurements. It was issued under the Federal Acquisition Supply Chain Security Act of 2018, which provides a framework for addressing certain security risks in federal technology supply chains.

Agencies can seek exceptions through the applicable process, but the default requirement is to remove covered technology and prevent new purchases.

The order does not amount to a general prohibition on all AI models developed in China. Its stated scope concerns DeepSeek, affiliated entities and third-party products covered by the directive.

Why Is Washington Concerned About DeepSeek?

The central concern identified by DHS is the potential exposure of sensitive information when government employees or vendors use DeepSeek technology.

The department says the models present multiple potential security risks and warns that sensitive data could be transferred to China. Its reasoning includes concerns about the location of DeepSeek’s infrastructure and the extent to which Chinese law may allow authorities to access company-held information.

For government agencies, the concern is not limited to information that employees deliberately enter into an AI chatbot. Depending on how a tool is integrated into an organisation’s systems, an AI application could interact with documents, internal workflows, software repositories or other information.

If a service sends prompts or related data to an external provider, agencies need to understand what information is transmitted, where it is processed, how long it is retained and who can access it.

These are general considerations in AI procurement, rather than proof that every deployment of DeepSeek transfers sensitive information to the Chinese government. The DHS order sets out the government’s security rationale; it does not establish that every DeepSeek implementation has been compromised.

Why hosting and deployment matter

DeepSeek’s models can be accessed through hosted services and, depending on the model and licence, deployed on infrastructure controlled by other organisations.

Those deployment options can produce different security arrangements. Using a provider’s hosted service may involve sending information to that provider, while running a model within an organisation’s own environment can change where data is processed.

However, local deployment does not automatically make a model secure. Organisations must still assess software vulnerabilities, access controls, model behaviour, system integrations and the risks associated with the data used by an application.

The DHS order is broader than a rule focused only on the location of servers. Its inclusion of third-party products developed using DeepSeek implementations could complicate decisions about which systems must be removed.

Why the Third-Party Restriction Matters

The third-party provision could have consequences for government contractors and software vendors that incorporate DeepSeek technology into their products.

For example, an agency might purchase an enterprise software service that includes an AI feature supplied by an external vendor. If that feature uses a covered DeepSeek implementation, the agency may need to determine whether the product falls within the order.

This creates a supply-chain identification challenge. Government IT departments need visibility into the models used by their suppliers, the services those models support and whether the relevant components fall within the directive’s scope.

The wording could also create uncertainty about how to classify software that was developed with assistance from DeepSeek but does not use a DeepSeek model at runtime. The order refers to products and services developed using DeepSeek implementations, and the precise application of that language may require further guidance.

Technology vendors may need to review their products, document their AI dependencies and communicate with government customers about compliance.

The issue demonstrates how AI procurement increasingly involves more than evaluating a product’s features and price. Agencies also need to assess where its models come from, how they are deployed and which third-party components they depend on.

Is DeepSeek Banned Across the United States?

No. The directive is specifically aimed at federal civilian agency systems and procurement. It does not, by itself, prohibit ordinary Americans, private companies or other non-federal organisations from using DeepSeek.

The US government has previously restricted DeepSeek in certain government environments, including national security-related systems and individual agency networks. The new order expands the restrictions across federal civilian IT and procurement under the specified framework.

The distinction matters because a government supply-chain restriction and a nationwide consumer ban have different legal scopes and consequences.

Private businesses remain responsible for assessing their own security obligations, industry rules and contractual requirements. A company handling sensitive information may decide that a particular AI service is unsuitable even when no government-wide prohibition applies to it.

Similarly, a model being available for local deployment does not automatically mean it is approved for government use. Federal agencies must follow the procurement and security rules that apply to their systems.

China’s Response to the DeepSeek Restrictions

China has strongly criticised the US decision. Its commerce ministry said the restrictions harm Chinese companies’ interests and urged Washington to revoke what it described as discriminatory measures.

Beijing also argued that the move conflicts with understandings reached by Chinese President Xi Jinping and US President Donald Trump. Chinese officials have called for the two countries to cooperate through their AI dialogue mechanism rather than politicise technology and trade.

The dispute highlights the challenge of managing AI security while maintaining commercial and diplomatic engagement between the world’s two largest economies.

Washington’s position is that government systems require safeguards against potential exposure of sensitive information. Beijing’s position is that the restrictions unfairly target Chinese technology companies.

These are competing government positions. The existence of the order does not independently settle every technical question about DeepSeek’s security, just as China’s objections do not remove the compliance requirements imposed on US civilian agencies.

The wider relationship between the two countries will influence how AI products, computing infrastructure and technology supply chains are treated in the future.

Why DeepSeek Matters in the Global AI Market

DeepSeek attracted global attention in January 2025 when its R1 reasoning model demonstrated competitive capabilities while being associated with comparatively low development and operating costs.

Its emergence challenged assumptions about the resources required to develop advanced AI systems and increased investor scrutiny of spending on AI infrastructure. It also demonstrated how open-weight models could provide alternatives to proprietary systems offered by major US technology companies.

Open-weight models make trained model parameters available for download and use under the applicable licence. They can allow developers to modify, customise or deploy models on their own infrastructure, although the extent of permitted use depends on the specific licence and accompanying terms.

This flexibility has made models such as DeepSeek attractive to developers and organisations seeking greater control over their AI systems or lower operating costs.

The same flexibility creates additional governance questions for large organisations. When a model can be integrated into many applications, a security review focused only on the original chatbot may not capture every deployment.

For federal agencies, the new order makes procurement compliance and technology provenance particularly important. For the broader market, it underscores how security policy can affect the adoption of AI models even when those models are commercially competitive.

What the Order Means for AI Companies and Government Contractors

US government agencies and their suppliers may need to review their AI inventories, procurement contracts and software dependencies to determine whether covered DeepSeek technology remains in use.

That work can involve identifying direct deployments, checking cloud services, consulting vendors and establishing replacement plans for affected systems. Contractors may also need to confirm which AI models power the products they provide to federal customers.

For AI companies competing for government business, the order highlights the importance of security documentation, transparent deployment arrangements and clear information about model dependencies.

US-based providers could benefit if agencies move to alternative products, but the restrictions also introduce costs for government departments and vendors that must replace or reconfigure existing systems.

The effect will depend on how much DeepSeek technology agencies currently use, how the third-party provision is interpreted and which alternatives meet their technical and budget requirements.

A change in approved vendors does not automatically guarantee that replacement systems will be cheaper, more capable or free of security vulnerabilities. Agencies will still need to evaluate products against their operational needs and security standards.

The Bigger Picture

The DeepSeek order shows how AI is becoming a central issue in government technology procurement and national security policy. Authorities are no longer evaluating only standalone applications; they are increasingly considering the models and dependencies embedded throughout software supply chains. This approach may help agencies address potential data-exposure risks, but broad restrictions can also create implementation challenges when third-party products incorporate AI in less visible ways.

The decision also reflects the wider US-China technology rivalry. DeepSeek’s low-cost, open-weight approach has helped it compete internationally, while the US government has prioritised limiting potential security exposure in its own systems. The long-term implications will depend on how the order is implemented, how vendors respond and whether further guidance clarifies the treatment of third-party applications.

Looking Ahead

Federal civilian agencies and their contractors will need to identify covered DeepSeek products, remove them from affected systems and stop purchasing prohibited products and services. The exception process and the interpretation of third-party coverage will be important to implementation, particularly for cloud software and AI-powered applications that do not openly advertise their underlying models.

For DeepSeek and the wider AI industry, the order creates another regulatory barrier to government adoption in the United States. It does not establish a blanket restriction on private use, but it may influence how businesses assess data handling, deployment choices and AI supply-chain transparency. As AI becomes more deeply embedded in software, those questions are likely to remain central to technology procurement.

Frequently Asked Questions

1. Why has the US ordered agencies to remove DeepSeek?

The Department of Homeland Security cites national security risks, including the potential exposure of sensitive government information to Chinese authorities when DeepSeek products are used by government employees or vendors.

2. Which organisations are covered by the order?

The order applies to federal civilian agencies and their covered IT procurement and systems. It also includes DeepSeek-affiliated products and third-party products and services developed using covered DeepSeek implementations.

3. Is DeepSeek banned for everyone in the United States?

No. This directive does not create a nationwide ban on DeepSeek for private individuals or businesses. Its stated scope concerns federal civilian agency systems and procurement.

4. Can federal agencies request exceptions?

Yes. Agencies can seek exceptions through the applicable process, subject to the requirements established for the order.

5. What has China said about the ban?

China has condemned the restrictions, arguing that they discriminate against Chinese companies and politicise technology and trade. Beijing has called for the measures to be revoked.

6. Does the order prove that DeepSeek sends all user data to China?

No. The order states the US government’s security concerns and the potential for sensitive data exposure. It does not establish that every deployment of DeepSeek transfers all user data to Chinese authorities.