The US Department of Homeland Security has ordered all artificial intelligence technology from the Chinese firm DeepSeek to be removed from most government systems, according to BBC News. The notice, made public on Monday, bans all use of DeepSeek within civilian government agencies and any technology built with the aid of its AI models. The ban took effect from the end of July, even though it was only disclosed this week.
The scope is narrower than a blanket federal prohibition but wider than a ban on a single app. BBC News reports that the order covers civilian agencies including the Department of the Treasury, the Department of State and the Department of Justice, while agencies within the Department of Defense are excluded. Decrypt reports that the order also tells agencies to stop buying DeepSeek and any outside service built on top of it, extending the restriction to third-party vendors rather than the chatbot alone.
What the Notice Actually Says
The two accounts differ in emphasis. BBC News notes that the DHS notice did not detail the exact security concerns surrounding DeepSeek. Decrypt, by contrast, reports that DHS says the models pose "multiple threat vectors", meaning several separate ways things can go wrong, and that using them "may result in the transfer of sensitive data to the PRC" if government employees or vendors use them, potentially producing "actionable intelligence on U.S. persons and/or government agencies." The distinction matters for anyone reading the order as a compliance document. The published rationale, as reported by Decrypt, is a data-flow and intelligence-risk argument rather than a technical vulnerability finding. That framing points procurement teams towards where the exposure sits: not only in DeepSeek's own application but in any product that has embedded its models.Why Open Source Does Not Settle the Question
DeepSeek's models are open source, which BBC News describes as free to access and less expensive to use, and the company has been seen as a disruptive force in the AI technology race. Decrypt notes that the model can run locally or be hosted by any provider, not just a China-based one, citing Perplexity's US-located servers as an example. DeepSeek's developers do, however, host a version on servers in China for users of its own service. That combination is what makes the order awkward to implement. An open-weight model can sit inside a domestic product with no obvious Chinese hosting dependency, yet still fall within a clause covering technology built with the aid of DeepSeek models. BBC News reports that companies including Snowflake, Vercel and OpenRouter incorporate DeepSeek in their offerings to customers, and that startups have flocked to open source models like DeepSeek because using them costs a fraction of a closed model such as Google's Gemini, Anthropic's Claude or OpenAI's Codex.The Cost Argument That Built DeepSeek's Position
The commercial logic behind that adoption is stark. BBC News reports that DeepSeek said last year it took about $6m to train the debut version of its model, while the costs associated with OpenAI's GPT-4 were roughly $100m. Decrypt reports that the company became a household name on 27 January 2025, when its open-source model, built for under $5 million, knocked Nvidia's stock down 16% and erased about $600 billion of the chipmaker's value, with US tech as a whole losing an estimated $1 trillion that day. The order lands as DeepSeek keeps shipping. Decrypt reports that the preview of its V4-Pro model, released in April, is priced about 98% below OpenAI's GPT-5.5 Pro and is free to download and modify under an MIT license, and that an updated version of its V4.1-Flash model was released this week. BBC News reports that the most recent version was released last week and that users have praised it online for beating advanced OpenAI tools on common tasks such as software coding.A Wider Federal Push
The DHS order is one step in a broader sequence. Decrypt reports that a July DHS memo, made public on Monday, had already told agencies to monitor traffic flowing in and out of government websites for DeepSeek-related risks, and that the Commerce Department and the Navy previously banned DeepSeek on government devices, with New York, Tennessee and Virginia taking similar steps. Decrypt also reports that the Federal Communications Commission votes on 29 October on barring all Chinese labs from testing electronics for US use. BBC News places the order against criticism of Chinese AI from the Trump administration and from American firms. It reports that White House tech lead Michael Kratsios wrote earlier this year in an internal memo that "malicious exploitation" of American AI firms had been occurring, mainly by rival firms in China, and that Anthropic and OpenAI have claimed their AI models were essentially copied by some Chinese AI firms, including DeepSeek. Decrypt reports that in April 2025 a House select committee on the Chinese Communist Party alleged that DeepSeek secretly sends American user data to the Chinese government, and that the same committee said DeepSeek may have used about 60,000 Nvidia chips, including 20,000 restricted under US export controls.Commercial Impact
The immediate burden falls on vendors selling into US civilian agencies. Any supplier whose product embeds DeepSeek models, whether directly or through a downstream component, now needs to establish whether its government customers are covered by the order and what the clause on technology "built with the aid of" those models captures. The exclusion of Department of Defense agencies creates a split map of exposure across federal buyers rather than a single clean line. For DeepSeek, the direct revenue at stake in US civilian government use is unlikely to be the main issue, given that its models are free to access and widely distributed through third parties. The more consequential effect is on the calculus of Western software firms that chose open-weight Chinese models on cost grounds. A procurement restriction in the largest public-sector market gives those vendors a reason to document model provenance and offer substitutes, even where their commercial customers face no such rule.Risks and Unknowns
Several points remain unresolved. The DHS, DeepSeek and the Chinese embassy in Washington DC were contacted for comment by BBC News, and Decrypt reports that the Chinese Embassy did not immediately comment. Neither account establishes how agencies will verify that third-party services are free of DeepSeek-derived components, nor what evidence vendors must supply. The security rationale is also only partly public. BBC News states that the notice did not detail the exact security concerns, while Decrypt quotes the threat-vector language. Whether the order is enforced through procurement conditions, traffic monitoring or self-certification is not set out in either report. And the timing is unusual: a ban effective from the end of July, disclosed in October, leaves a gap during which compliance status was not publicly known.What to Watch
The FCC vote on 29 October will indicate whether the US is moving toward a broader exclusion of Chinese AI testing and components, as Decrypt reports. For vendors, the practical question is narrower and more immediate: which products in a government-facing catalogue contain DeepSeek models or were built using them, and what documentation can be produced if an agency asks.Sources and References
- BBC News (bbc.co.uk) — US bans DeepSeek AI from government use as tech tension with China continues
- Decrypt (decrypt.co) — US Orders Civilian Agencies to Rip DeepSeek Out of Government Systems
