The United States has ordered the removal of all artificial intelligence technology from Chinese firm DeepSeek from most government systems. The Department of Homeland Security (DHS) has banned the use of DeepSeek within civilian government agencies, as well as any technology built with the aid of its AI models.
The order was published on Monday, but the ban has been in effect since the end of July. It applies to civilian agencies, including the Department of the Treasury, the Department of State and the Department of Justice. Agencies within the Department of Defence are exempt from this measure. The DHS notice did not detail the specific security concerns regarding DeepSeek.
The ban follows criticism of Chinese AI by the Trump administration and American firms. White House technology chief Michael Kratsios wrote in an internal memorandum earlier this year that 'malicious exploitation' of American AI firms was taking place, primarily by rival firms in China. The White House has criticised Chinese AI firms this year, and US President Donald Trump has made it clear that he views China as a key American rival in AI development, which he considers crucial for the country's future.
Anthropic and OpenAI, both of which are preparing to list on the public stock market in the coming months, have also claimed that some Chinese AI firms, including DeepSeek, have essentially copied their AI models. DeepSeek is a powerful open-source AI model, meaning it is free to access and cheaper to use, and is considered a disruptive force in the current AI technology race.
DeepSeek's popularity surged following its first public release in 2025, and the models were so in demand that the company's servers sometimes crashed. DeepSeek's free chatbot application is one of the most downloaded apps on the Apple App Store rankings. American tech firms also use DeepSeek's AI models — for example, Snowflake, Vercel and OpenRouter include DeepSeek in their user offerings, and startups have rushed towards open-source models like DeepSeek.
For a technology or AI developer, using DeepSeek in their work costs a fraction of what a closed model like Google's Gemini, Anthropic's Claude or OpenAI's Codex would cost. The cost DeepSeek incurred to develop its models was astronomically lower than what it cost American AI firms to develop theirs. Last year, DeepSeek said it took around $6 million (£4.5 million) to train the first version of its model, while costs associated with OpenAI's GPT-4 were approximately $100 million. Such revelations sent technology stocks into a decline.
The latest version of DeepSeek was released last week. Users of the model praised it online, saying it beats even the most advanced AI tools from OpenAI in common AI tool tasks such as computer programming.
