Washington has widened its push against Chinese artificial intelligence to the software running inside its own agencies. The latest step is an exclusion and removal order from the Department of Homeland Security. Although it was announced only on Monday, the ban has bound civilian departments including Treasury, State and Justice since late July, with Defense Department agencies left outside it, the BBC reported. It also reaches third-party products built with DeepSeek's models, not just the company's own apps, the notice states.
The department set out several reasons for acting. The company's servers are located in China, where the legal system gives state cyber and intelligence agencies extensive access to private infrastructure, and the models themselves could carry malware, the order says.
DeepSeek became a household name because of its price tag. The company said it spent about $6m training its debut model, compared with roughly $100m for OpenAI's GPT-4, a disclosure that sent technology stocks tumbling last year, according to the BBC. Its open-source models have since been adopted by startups and by US firms including Snowflake, Vercel and OpenRouter, and its latest version appeared only last week. The order lands more than a year after OpenAI and Anthropic, along with federal officials, accused DeepSeek of systematically distilling their models, CyberScoop noted.