Homeland Security Orders Removal of Chinese AI from Federal Systems

The Department of Homeland Security has directed federal civilian agencies to eliminate AI technologies developed by China’s DeepSeek from their systems. This move comes as a significant setback for a Chinese model that has been gaining traction recently.

According to a notice from the DHS, all products related to DeepSeek, including third-party services that utilize its technology, must be removed from civilian agency IT systems. Additionally, any future procurement of DeepSeek’s products and services is prohibited.

The notice specifies that the Chinese company’s AI models “present multiple threat vectors.” It warns that using these products “may result in the transfer of sensitive data to the [People’s Republic of China]” if they are employed by U.S. government employees or vendors.

This DHS directive falls under the Federal Acquisition Supply Chain Security Act of 2018. It follows a review led by the Federal Acquisition Security Council, which provided recommendations backed by evidence. However, the announcement lacks details about the specific security concerns that led to this decision, and there’s no set timeline for agencies to comply.

Founded by Liang Wenfeng, DeepSeek is among a number of Chinese AI models that offer more economical alternatives to products from major U.S. companies like Anthropic and OpenAI. While U.S. models still lead in performance, the disparity has narrowed somewhat over the last year.

DeepSeek first made headlines early last year with its lower-cost AI models, which had a notable impact on major U.S. tech stocks, pushing Nvidia, Microsoft, and Tesla into what can only be described as a dramatic $1 trillion drop in value.

DeepSeek is known for its open-weight model, which allows for easier customization compared to the closed models from companies like Anthropic, Google, and OpenAI. The term “weights” refers to trained parameters, including the model’s source code and training data, which remain confidential in closed models but are accessible in open models.

The previous Trump administration had grown increasingly cautious regarding China’s rapid advancements in AI and the potential for Chinese companies to appropriate domestic AI technology.

Last year, the Commerce Department banned DeepSeek from government devices, and the U.S. Navy instructed its personnel to avoid using the model due to possible security risks.

Moreover, New York State has also prohibited the use of DeepSeek on government devices and networks, citing “serious concerns” about data privacy and censorship risks associated with the application.