US Orders Federal Agencies to Remove DeepSeek AI Products

Concerns over government data security have prompted the United States to expand restrictions on Chinese artificial intelligence company DeepSeek, ordering civilian federal agencies to remove its products and software built using its models. The Department of Homeland Security (DHS) issued the directive to prevent potential exposure of sensitive government information to China, with the restrictions also prohibiting new purchases of covered products and services.

According to reports published by Clubic and TradingView on October 20, 2026, the order was dated July 24 and became public on October 19. It covers DeepSeek, its affiliates, and third-party software incorporating its technology, including products that do not visibly identify DeepSeek as their underlying model provider.

The DHS cited potential transfers of sensitive information to China, possible exploitation of government data for intelligence purposes, and security vulnerabilities. TradingView also reported that the department referenced vulnerabilities identified in DeepSeek's iOS application. The concerns cited by US authorities do not establish that every product using DeepSeek technology transmits information to Chinese authorities.

The restrictions apply to civilian federal government systems rather than all American users. Private individuals and companies outside the scope of the directive can continue using DeepSeek, while an exemption procedure allows covered agencies to seek waivers under the order's provisions.

DeepSeek's downloadable model weights complicate enforcement because organizations can run the technology on their own servers without connecting directly to the company's infrastructure. According to Clubic, third-party developers can incorporate the models into their own software, making the technology's origins less apparent to users and administrators.

The company initially attracted international attention with its V3 and R1 models, which challenged established competitors in generative AI. Clubic reported that V3 uses a mixture-of-experts architecture with 671 billion parameters, activating only a portion of them during individual computational steps. R1 gained attention for reasoning, mathematics, and programming tasks.

US restrictions on DeepSeek build on earlier measures affecting government departments, military organizations, and several states. The Department of Commerce and the US Navy were among the entities that had already restricted its use on government devices, while New York, Tennessee, and Virginia also imposed restrictions, according to Boursorama.

Other governments have adopted similar precautions. Australia banned DeepSeek from government systems and devices in February 2025, while the Czech government prohibited public administration from using the service in July 2025 over data security concerns. Taiwan also barred government services from using DeepSeek in February 2025, citing security risks and concerns about censorship and potential data transfers to China.

European authorities have examined the company's privacy practices. France's data protection authority sought information about how the system operates and its potential privacy risks in January 2025. In June 2025, Germany's data protection commissioner said Apple and Google had been asked to remove DeepSeek from their online app stores over data security concerns.

Italy's competition authority, the AGCM, closed an investigation after DeepSeek accepted binding commitments concerning disclosures that its AI system could generate false information. The authority had previously blocked the application over insufficient information about its handling of personal data. The Dutch privacy regulator also announced an investigation into DeepSeek's data collection practices in January 2025 and urged users to exercise caution.

South Korea suspended new downloads of DeepSeek in mid-February 2025 after the company acknowledged failing to account for certain personal data protection requirements. The service became available again in late April 2025, although a government ministry had temporarily blocked employee access over security concerns.

India's Finance Ministry instructed employees in early February 2025 to avoid using AI tools, including DeepSeek and ChatGPT, for official work because of risks involving confidential documents and government data. Russia took a different approach: in early February 2025, President Vladimir Putin instructed Sberbank to collaborate with Chinese researchers on joint AI projects, according to a senior executive at the country's largest bank cited by Reuters.

According to Boursorama, the Trump administration has also considered restrictions that could prevent DeepSeek from purchasing US technology and has debated whether to prohibit Americans from accessing its services. In December, nine US lawmakers wrote to Defense Secretary Pete Hegseth urging the Pentagon to add several Chinese technology companies, including DeepSeek, to a list of entities suspected of assisting China's military. Seven Republican senators had separately asked the Commerce Department in August to assess potential data security vulnerabilities associated with Chinese open-source AI models.