Blocking a single application on government computers may sound straightforward. Finding every place that application or its underlying technology is used is a different challenge. With artificial intelligence, a model can be embedded in business software, cloud services or specialized tools without appearing under its original name. That is what makes the latest US decision particularly significant. On October 19, the Department of Homeland Security made public an order excluding DeepSeek technology from federal civilian agency information systems. It also bars covered government procurement of the company's products and services. The restrictions do not stop at DeepSeek's consumer chatbot. They encompass affiliated entities and third-party products developed using DeepSeek implementations. For federal IT departments, compliance may therefore require more than removing a recognizable app. They will need to identify AI components embedded in existing services. The order treats artificial intelligence as a supply-chain security concern, potentially affecting software contracts, vendor assessments and system inventories.

A July order made public in October

The timeline contains an important distinction. The public notice filed on October 19 is not necessarily the date on which the prohibition was originally decided. The official document includes Exclusion and Removal Order 2026-01, bearing an original release date of July 24, 2026. It was signed by Homeland Security Secretary Markwayne Mullin. The notice became available for public inspection on October 19, with formal publication in the Federal Register scheduled for October 20. This gap explains why a measure carrying a July date became a major news story several months later. DHS issued the order under the Federal Acquisition Supply Chain Security Act of 2018. That law provides a framework for addressing risks from technology products and services used in government procurement and information systems. The action followed consideration of a recommendation from the Federal Acquisition Security Council. This is not a newly enacted law prohibiting all Chinese artificial intelligence. It is a targeted administrative order affecting specified organizations and technologies.

Two obligations: stop buying DeepSeek and remove existing products

The order has two main requirements. First, federal civilian agencies must exclude DeepSeek and its covered affiliates from relevant IT procurement. That obligation extends to approved-product lists, certain equipment catalogs, online purchasing channels and other government procurement mechanisms. Second, agencies must remove covered DeepSeek-related products from existing information systems. The removal requirement also includes third-party products and services developed using DeepSeek implementations. That second element may prove more complicated. An agency can usually identify employees who directly use the DeepSeek chatbot. It may be much harder to identify business applications whose AI features are powered by models developed by the Chinese company. The model name may never appear in the user interface. Compliance could therefore require vendor questionnaires, software inventories and reviews of existing contracts. The order is concerned with the technology underlying a product, not just the brand name shown to customers. That approach reflects the increasingly complicated supply chains surrounding modern AI software.

Why the US government considers DeepSeek a security concern

DHS identifies several categories of risk. The first involves the handling of data. DeepSeek is based in China, where national security legislation can create obligations for domestic companies. US officials argue that those conditions raise concerns about information processed by the provider's services. The order also discusses personal information and technical metadata that may identify or track users. Another concern involves the content of prompts submitted to AI models. When a government employee asks a chatbot to summarize a document or analyze an operational issue, the prompt may contain information that should remain within the agency. If that information is processed using infrastructure in another country, its confidentiality becomes a national security concern. DHS argues that seemingly separate pieces of information may become more revealing when combined with other sources. The risk therefore extends beyond documents explicitly classified as secret. Routine administrative data can reveal details about government programs, personnel and organizational activities. These concerns explain the government's decision. They do not prove that every DeepSeek session has resulted in unauthorized disclosure.

Potential vulnerabilities are not proof of widespread compromise

Cybersecurity is another element of the order. DHS raises the possibility of malware or other security vulnerabilities in covered products and models. It also references earlier research identifying security and privacy flaws in the DeepSeek iOS application. Researchers reported several issues involving the mobile software in 2025. Those findings contributed to the scrutiny surrounding the service. However, three distinctions are essential. A vulnerability identified in one version of an application is not necessarily present in every later version. The possibility that a product may contain malicious components is not evidence that every implementation does. And concerns about data transfer do not establish that a widespread government compromise has occurred. The order is a supply-chain risk management decision. It is not a public forensic report demonstrating successful intrusions into all agencies that have used DeepSeek. The administration has decided that the identified risks are unacceptable for the systems covered by the order. Such a preventive decision does not require public proof that every deployment has already been breached.

Third-party applications create a difficult compliance challenge

DeepSeek became widely known partly because of the accessibility of its AI models. Developers can use its hosted services. Some models also have openly available weights, allowing organizations to deploy them on their own infrastructure. That flexibility encourages integration into other software. A company might build a document assistant, coding tool or search feature using a DeepSeek model. The customer may never know which AI technology powers the feature. The US order complicates that arrangement. It expressly covers certain third-party products and services developed using DeepSeek implementations. A vendor seeking to sell software to a federal civilian agency must therefore understand whether its technology falls within the restrictions. Public-sector customers will need similar visibility. A contract with an American software supplier does not automatically mean that every underlying component was developed in the United States. The origin and composition of AI models are becoming procurement compliance questions. That change could affect how vendors document and disclose the technologies they integrate into their products.

Running a model locally is different from using a Chinese-hosted service

There is an important technical distinction between hosted and locally deployed AI. When someone uses a service operated directly by DeepSeek, requests are processed by the provider's infrastructure under its applicable terms. Data location and handling requirements must be evaluated accordingly. A model with downloadable weights can also be deployed on servers operated by another organization. In that case, user prompts do not necessarily follow the same path. Running an AI model locally does not automatically mean that every conversation is transmitted to DeepSeek. However, this architectural distinction does not necessarily exempt a product from the DHS order. The official text extends to certain third-party implementations. It would therefore be incorrect to assume that a DeepSeek model becomes permissible for federal civilian use simply because it runs on an American server. Each covered organization must consider the actual rules and the specific product. The broader implication is that government technology policy increasingly addresses not only where data travels but also which technologies are embedded in sensitive systems.

Federal agencies may request exceptions

The order includes a process for seeking exceptions. Covered agencies may submit a written request for relief from specified requirements. The document provides a sixty-day period following issuance of the order for submitting those requests. That should not be confused with a new blanket sixty-day permission to keep using DeepSeek after the order became public in October. An exception request must describe the product involved and explain the precise relief being sought. It must also provide a compelling justification. The order identifies mission-critical functions and certain national security considerations as examples of relevant issues. Agencies must describe alternative measures they would use to address the identified risks. The Cybersecurity and Infrastructure Security Agency participates in reviewing exception requests and providing technical assistance. The mechanism recognizes that replacing software dependencies may sometimes be difficult. However, an exception is not automatic. The relevant authorities must examine the request and decide whether the proposed arrangement is acceptable.

DHS encourages stronger monitoring during removal

The order does more than direct agencies to uninstall software. While covered products remain in government systems, it encourages additional cybersecurity measures. These include monitoring inbound and outbound network traffic. The document also recommends enhanced logging and controls designed to identify potential data exfiltration. Agencies are encouraged to detect and flag anomalous activity. Those recommendations are consistent with supply-chain incident management. When an organization determines that a technology component should no longer be used, it also needs to know where the component is installed and what access it has. However, recommending stronger monitoring does not mean authorities discovered data theft in every affected agency. The measures are designed to manage potential exposure during the removal process. They may also require organizations to improve their inventories of AI applications. For federal IT teams, that operational work could be as challenging as the policy decision itself.

The order does not ban private DeepSeek use across America

The scope is another important point. The order applies to covered information systems and procurement activities of federal civilian executive branch agencies. It does not mean a private individual in the United States is automatically breaking the law by using DeepSeek on a personal smartphone. Nor does the order itself prohibit every private American business from accessing the company's models. Organizations may nevertheless choose to adopt similar restrictions through their own security policies. Software vendors supplying federal agencies may also need to modify products to meet the new requirements. The document distinguishes DHS authority over civilian systems from the responsibilities of other federal authorities covering certain defense and intelligence systems. It would therefore be inaccurate to treat this single order as the complete regulatory framework for every US government organization. Its immediate effect concerns federal civilian IT systems, procurement and relevant suppliers. Any broader effects will depend on additional decisions by other organizations.

DeepSeek was already restricted in parts of government

The decision did not emerge without precedent. Since DeepSeek gained international attention, some government organizations have raised concerns about its use on official devices. In the United States, the Commerce Department and the Navy had previously introduced restrictions in certain environments. States including New York, Tennessee and Virginia had also taken targeted action. Other countries have imposed their own government-system limitations. The DHS order now provides a broader federal civilian framework for these concerns. The development occurs amid growing technological competition between the United States and China. However, it should not be reduced entirely to commercial rivalry. For public administrations, AI models are software components capable of processing sensitive information. Their origin, hosting arrangements and maintenance practices can therefore matter in the same way as other critical technologies. AI is becoming more deeply integrated into public-sector infrastructure security policy.

Software suppliers may face new transparency demands

Companies selling technology to US federal agencies may face closer questions about their AI dependencies. A document search service might include an embedded language model. A coding assistant may use several providers. An enterprise application might change its underlying model without immediately making that change obvious to end users. Those practices become difficult when a customer prohibits specific technologies. Suppliers need to understand precisely what their own products contain. They may have to disclose which models process requests, where data is handled and which technical dependencies are involved. In some cases, replacing a model may be straightforward. In others, it could affect output quality, compatibility or operating costs. Government agencies will also need better inventories. A conventional list of installed applications may no longer capture the AI components used inside cloud-based software. That is one of the broader lessons of the DeepSeek decision. Managing artificial intelligence security increasingly requires visibility into software supply chains rather than merely controlling which chatbots employees open.

National security meets competition in artificial intelligence

The order inevitably has geopolitical implications. DeepSeek operates in a sector where Chinese companies increasingly compete with major American AI providers. Its open-weight models have helped expand access to capable and relatively affordable AI systems. That accessibility encouraged adoption by developers around the world. For Washington, the spread of those models raises new questions when they become components of government technology. DHS has taken a risk-management approach. The order is not a ranking of DeepSeek's capabilities against those of OpenAI, Anthropic or Google. A model may perform well and still be considered unsuitable for a sensitive government environment. Answer quality alone does not determine whether a technology satisfies public-sector security requirements. Contractual protections, data control and legal access to infrastructure also matter. At the same time, the government's risk assessment must be distinguished from publicly proven technical incidents. The concerns cited by DHS explain the administration's decision. They do not establish that every DeepSeek deployment has already caused harm.

Open-weight models create a more complicated policy debate

The case also illustrates a tension surrounding open-weight AI. Making model weights available allows organizations to modify and deploy systems on their own servers. That can reduce dependence on external hosting services. It can also support technical review and customization. But access to model weights does not automatically guarantee security. Software components, deployment practices and model behavior still matter. The American order adds a regulatory criterion related to the origin of certain implementations. An organization may control its own infrastructure without necessarily being permitted to use a particular model in a federal civilian system. That distinction could influence public procurement decisions. IT departments may need to assess model capabilities, data handling and legal restrictions together. Open-weight AI is not inherently excluded from government use by this development. Its deployment increasingly depends on the applicable rules and the specific technology involved.

A decision with lasting implications for government procurement

DHS says the order will be reviewed within one year and annually thereafter. That means the measure is subject to a defined review process. Future technical information and changes in the risk environment may be considered through that framework. For DeepSeek, the immediate consequence is clear. Its covered products and implementations are excluded from an important segment of US federal civilian procurement and information systems. For third-party suppliers, the implications may be less obvious. They must determine whether their products contain implementations falling within the restrictions when selling to covered agencies. For government IT departments, the practical challenge involves removing affected technology, reviewing exception requests and improving visibility into software dependencies. The decision reflects a broader transformation. Artificial intelligence is becoming part of the information technology supply chain, alongside operating systems, networking equipment and cloud services. AI models can process sensitive data, influence decisions and operate within professional software. The US restrictions on DeepSeek show governments increasingly applying structured security requirements to those models. The difficult part will be implementing such policies precisely while keeping potential risks, demonstrated incidents and the actual legal scope of the rules clearly separated.