US Orders Federal Civilian Agencies to Purge DeepSeek AI and Derived Tools

The United States government has formally ordered federal civilian agencies to purge artificial intelligence software developed by Chinese developer Hangzhou DeepSeek Artificial Intelligence Co. from their internal networks and immediately halt all future purchases. Made public in an official Federal Register filing, the directive covers not only DeepSeek branded applications and web interfaces, but also reaches into the broader software supply chain by barring third-party products and commercial services built upon DeepSeek implementations^{[1, 3]}.

The aggressive federal measure comes amid an intensifying global technology competition between Washington and Beijing. DeepSeek stunned Western tech leaders and financial markets in early 2025 by deploying reasoning models like DeepSeek-R1 that achieved high performance at a fraction of the compute cost demanded by American competitors. By invoking federal supply chain statutes rather than temporary executive advisories, the administration has established a sweeping, permanent standard that treats open-weight Chinese foundation models as an institutional security vulnerability.

Statutory Grounding and Operational Scope

The action, designated as Exclusion and Removal Order #2026-01, was issued under the Federal Acquisition Supply Chain Security Act of 2018 following a risk evaluation by the Federal Acquisition Security Council. According to Bloomberg and CyberScoop reporting, the underlying order was signed by Homeland Security Secretary Markwayne Mullin in late July before being placed on public inspection and published in the Federal Register^[1].

The operational mandate applies across federal civilian executive-branch agencies, including the Department of the Treasury, Department of State, and Department of Justice. It excludes Department of Defense networks and dedicated intelligence-community systems, which manage foreign tech exposure under distinct national security directives. Agencies have a 60-day window to apply for exceptions directly through the Cybersecurity and Infrastructure Security Agency (CISA), but approvals require compelling justification demonstrating mission-critical need.

The directive outlines concrete enforcement measures that agencies must take regarding their procurement pipelines and digital infrastructure^[8]:

• Immediate removal of all DeepSeek installations, standalone software, and application programming interfaces (APIs) from agency hardware and cloud environments.
• A total ban on federal acquisition or renewal of contracts containing DeepSeek products, including removal from agency-approved software catalogs and federal e-commerce marketplaces.
• The identification and uninstallation of vendor-supplied, third-party software tools that rely on DeepSeek implementations^[3].
• Interim monitoring protocols requiring agencies to analyze data traffic, bolster exfiltration logs, and flag anomalous network interactions until all instances are stripped.

National Security Rationale and the Open-Weight Dilemma

In the official documentation, the Department of Homeland Security asserted that the foreign AI systems introduce multiple vectors of risk into federal infrastructure. Because DeepSeek maintains servers in China, officials warned that Chinese national intelligence and cybersecurity statutes give Beijing sweeping legal authority to compel companies to surrender enterprise data^[14].

DeepSeek's models present multiple threat vectors. DeepSeek servers are located in the PRC, and the PRC legal system permits extensive authority to government cyber and intelligence functions to access private company infrastructure.

Department of Homeland Security, Federal Register filing

Officials also expressed concern that account registration systems collect personally identifiable information and user-tracking data that foreign actors could exploit to fingerprint federal personnel. Additionally, regulatory evaluations from the National Institute of Standards and Technology flagged that DeepSeek models displayed higher susceptibility to prompt injections and jailbreaks than competing domestic systems.

Yet cybersecurity researchers point to a critical technical debate over how risk is evaluated. DeepSeek publishes open weights, enabling developers around the world to download models and run them entirely on local, disconnected hardware or sovereign cloud environments. Security professionals argue that a locally containerized model with air-gapped data pipes presents an entirely different threat posture than sending unencrypted corporate prompts to an external server in China^[15].

Supply Chain Complexities for Government Vendors

The derivative clause mandating the removal of third-party products built on DeepSeek implementations represents the directive's heaviest practical burden. Because open foundation models are readily adapted, modified, and integrated into complex enterprise systems, many federal contractors may not even realize that third-party microservices incorporate DeepSeek code or distilled weights.

Security analysts note that an AI software bill of materials does not traditionally log the intermediate models a developer tapped while testing or distilling code. Services such as automated AI routing gateways can also dynamically shift enterprise queries to various backend providers based on latency and cost, potentially routing federal data through DeepSeek models without explicit administrative intent.

Diplomatic Pushback and Strategic Friction

The formalization of the federal ban drew swift diplomatic criticism from Beijing. During a press briefing in Beijing, Chinese Foreign Ministry spokesperson Mao Ning said China would resolutely defend the legitimate rights and interests of its enterprises, arguing that Washington continues to overstretch the concept of national security to politicize regular scientific and commercial trade^[4].

The move follows prior piecemeal restrictions: the Pentagon blocked DeepSeek software in January 2025, while the Department of Commerce and the U.S. Navy separately prohibited personnel from using the platform on government-issued hardware. Individual states, including Virginia, New York, and Tennessee, previously enacted comparable bans on state devices. By shifting from ad hoc prohibitions to an organized exclusion under the Federal Acquisition Security Council, the federal government has firmly drawn its perimeter, forcing the enterprise tech sector to choose between Chinese open-weight software and lucrative federal contracts.