US bans DeepSeek AI from civilian government systems
The United States has ordered Chinese firm DeepSeek’s artificial intelligence technology removed from most government systems, banning its use across civilian agencies.
The Department of Homeland Security ban also covers technology built with the aid of DeepSeek’s AI models. It applies to civilian agencies, including the departments of the Treasury, State and Justice, but excludes agencies within the Department of Defense.
DHS made the order public on Monday, although the ban took effect at the end of July. Its notice did not detail the exact security concerns surrounding DeepSeek.
DHS, DeepSeek and the Chinese embassy in Washington have been contacted for comment.
The action follows criticism of Chinese AI companies by the Trump administration and American technology firms. White House technology lead Michael Kratsios wrote in an internal memo earlier this year that American AI companies had faced “malicious exploitation,” mainly by rival firms in China.
Anthropic and OpenAI have alleged that some Chinese AI companies, including DeepSeek, effectively copied their models. Both American companies are expected to list on public stock markets in the coming months.
President Donald Trump has identified China as the United States’ main rival in AI development, which he views as central to the country’s future.
DeepSeek has grown in popularity since its initial public release in 2025. Demand for its models has at times caused the company’s servers to go offline.
Its free chatbot is among the most downloaded apps in Apple’s App Store rankings. American technology companies Snowflake, Vercel and OpenRouter also incorporate DeepSeek into their customer offerings.
Startups have increasingly adopted open-source models such as DeepSeek, which are free to access and less expensive to use. For developers, using DeepSeek costs a fraction of using closed models such as Google’s Gemini, Anthropic’s Claude or OpenAI’s Codex.
DeepSeek’s model-development costs have also been far lower than those of American AI firms. Last year, DeepSeek said training the debut version of its model cost about $6 million, or 4.5 million pounds, compared with roughly $100 million associated with OpenAI’s GPT-4. The disclosure sent technology stocks falling.
DeepSeek released its most recent model last week. Users have praised it online, saying it outperforms even OpenAI’s most advanced tools on common AI tasks such as computer software coding.
Featured image credit
