DHS bans DeepSeek AI across civilian US government agencies

The US Department of Homeland Security has banned all use of DeepSeek's AI across civilian government agencies, with the order in force since the end of July and made public on Monday.

The directive covers DeepSeek's Chinese-built AI and any technology developed with the aid of its models. It binds civilian agencies including the Department of the Treasury, the Department of State and the Department of Justice. Agencies within the Department of Defense are excluded.

The DHS notice did not detail the specific security concerns behind the move. The BBC has contacted the DHS, DeepSeek and the Chinese embassy in Washington DC for comment.

Why does the ban matter commercially?

DeepSeek has become a disruptive force in the AI race because its models are open source — free to access and far cheaper to run than closed rivals like Google's Gemini, Anthropic's Claude or OpenAI's Codex. Startups have flocked to it.

The cost gap is stark. DeepSeek said last year it took about $6m (£4.5m) to train the debut version of its model. OpenAI's GPT-4 cost roughly $100m to develop. Those revelations sent tech stocks tumbling at the time.

The commercial footprint extends into the US itself. Snowflake, Vercel and OpenRouter all incorporate DeepSeek in their customer offerings. The company's free chatbot app ranks among the most downloaded in the Apple App Store.

What triggered the crackdown?

The ban follows sustained criticism of Chinese AI from the Trump administration and from American firms. President Donald Trump has framed China as the key US rival in AI development, a field he considers crucial to the country's future.

White House tech lead Michael Kratsios wrote in an internal memo earlier this year that "malicious exploitation" of American AI firms had been occurring, mainly by rival firms in China.

Anthropic and OpenAI — both poised to list on the public stock market in the coming months — have claimed some Chinese AI firms, including DeepSeek, essentially copied their AI models. Neither has publicly detailed the evidence behind those claims.

Can DeepSeek absorb the hit?

Demand for the models has repeatedly overwhelmed the company's own servers since its public release in 2025. Last week DeepSeek shipped its most recent version, and users online say it beats even OpenAI's most advanced tools on common tasks such as software coding.

The government order itself excludes the Pentagon and targets civilian systems, so the direct revenue impact on DeepSeek is limited. The signal to US corporate customers — Snowflake, Vercel, OpenRouter and the startups building on open weights — matters more: a federal security ban, even one without published specifics, gives procurement teams a reason to reconsider their cheapest supplier.