The Department of Homeland Security (DHS) has ordered federal civilian agencies to remove covered DeepSeek products from their IT systems and prohibit future purchases. According to Reuters, the restrictions also apply to third-party software and services developed using DeepSeek implementations.
The order could affect vendors that incorporate DeepSeek models into their products, even if customers interact only with the vendor's own software. Federal agencies may need to identify the underlying AI models used in the products they purchase and assess whether those implementations comply with government supply-chain security requirements.
DHS issued the order in July, before its October publication
The DHS order was originally issued on July 24, 2026, but its public notice was filed on October 19, ahead of publication in the Federal Register on October 20. The nearly three-month interval separates the order's issuance from its public disclosure.
Under the Federal Acquisition Supply Chain Security Act of 2018, the order requires federal civilian executive branch agencies to stop acquiring covered products and remove existing implementations, subject to exceptions. National security systems and sensitive compartmented information systems are governed by separate authorities. Private businesses and consumers are not subject to a nationwide DeepSeek ban.
Federal scrutiny of DeepSeek began before the order became public. Lawmakers introduced the No DeepSeek on Government Devices Act in February 2025, while some agencies had already restricted its use. The Commerce Department also confirmed restrictions on DeepSeek within its systems, according to CyberScoop.
DHS cites data-transfer risks and application vulnerabilities
DHS warns that DeepSeek-hosted services could expose sensitive information by collecting personal data and identifying users through device fingerprinting. The department also raises concerns that confidential information submitted through prompts could reach China, where it could be combined with public records to reveal information about U.S. individuals or government operations.
The order references an April 2025 investigation by the House Select Committee on the CCP. The committee expressed concerns that Chinese authorities could access personal information collected through DeepSeek under the country's data-access laws. These concerns form part of the committee's national-security assessment.
A separate February 2025 investigation by NowSecure identified vulnerabilities in DeepSeek's iOS application. Researchers found that the app transmitted certain registration and device information without encryption and used insecure methods for storing credentials. They also identified encryption weaknesses and data collection practices that could enable device fingerprinting.
These findings apply to the iOS app version examined in February 2025. Although DHS also raises the possibility of malware or other vulnerabilities within DeepSeek models, its published appendix does not provide evidence of malware embedded in a specific model. The documented app vulnerabilities do not prove that DeepSeek's underlying AI models contain malicious code.
Self-hosted DeepSeek models have different data exposures
DeepSeek's hosted services and independently deployed open-weight models involve different data flows. Through its DeepSeek-R1 repository, the company provides downloadable model weights that organizations can run on their own infrastructure instead of submitting prompts to DeepSeek-operated servers.
As LM Studio's documentation explains, local inference can also run offline after the necessary files are downloaded. This allows organizations to process prompts on their own systems, though the security of a deployment still depends on its software configuration and network connections.
NowSecure suggested self-hosting or alternative hosting as possible responses to the risks identified in DeepSeek's iOS app. Organizations using these arrangements may avoid sending prompts to DeepSeek's hosted service, but they remain responsible for evaluating the security of the downloaded model and the infrastructure running it.
Although independent hosting changes where information is processed, the DHS order provides no general exemption for self-hosted DeepSeek models. Agencies seeking exceptions must follow the applicable process, with CISA responsible for reviewing requests and providing technical assistance where appropriate.
During removal, DHS also recommends monitoring network traffic and maintaining enhanced logging and protections against data exfiltration. These measures address potential information transfers while agencies identify and remove affected implementations.
Third-party AI dependencies could affect federal procurement
Software vendors supplying federal agencies may face questions about which AI models power individual product features. For example, a platform that incorporates DeepSeek for document summarization or code generation could require review under the order even if the product is sold under another company's name.
Products that let customers choose among several AI models raise another implementation question. The published notice does not explain how agencies should evaluate software that includes DeepSeek as an available option but uses a different model by default.
Federal customers may therefore ask suppliers to identify which models are available in their deployments and demonstrate whether restricted options can be disabled. Vendors could also face questions about updates that introduce DeepSeek implementations into products previously approved for use. These are potential procurement implications rather than additional obligations explicitly described in the order.
The distinction between a software provider and its underlying AI suppliers could also influence private-sector purchasing decisions. Although the DHS restrictions apply to federal civilian agencies, businesses can examine model origins and hosting arrangements to understand which organizations process their information and maintain the computing infrastructure.
The order's third-party provision makes those dependencies relevant to federal procurement decisions, particularly when a commercial product can connect to multiple AI providers. The remaining uncertainty is how agencies will verify compliance across software that embeds DeepSeek models or makes them available as interchangeable options.
The DHS action could therefore require vendors to provide more information about the AI implementations inside their products. How consistently agencies apply those requirements will depend on the compliance procedures developed around the order.
