On Monday, 19 October 2026, the United States took its strongest step yet against one of the most closely watched artificial intelligence companies in the world. The Department of Homeland Security ordered civilian federal agencies to remove DeepSeek's AI products — and any third-party software built on DeepSeek models — from government IT systems, and banned those agencies from acquiring DeepSeek products or services going forward. The order, published in the Federal Register, marks a new peak in Washington's campaign against Chinese AI tools inside American government networks.

Background: DeepSeek's rise and Washington's suspicion

DeepSeek, a Chinese AI company, burst onto the global stage with models that rivalled the best Western systems while being developed at a fraction of the expected cost. Its releases forced a global conversation about whether America's lead in AI was as comfortable as assumed. But for US officials, the company's success came with a deep worry: a cutting-edge AI model from China, potentially interacting with American government data, looked like a security problem waiting to happen.

Washington's suspicion did not start with this order. The Department of Commerce and the US Navy had already banned DeepSeek from their systems, and several states — including New York, Tennessee and Virginia — took similar steps. Each of those moves treated DeepSeek as a special risk; what is new about the DHS order is its scale. It applies across the civilian agencies of the federal government — the departments and bodies that handle everything from public health and homeland security operations to benefits payments and environmental regulation.

What happened: the DHS order

According to reporting by Reuters' David Shepardson, and covered by outlets including Engadget and Gizmodo, the notice published Monday does two things. First, it directs civilian agencies to identify and remove DeepSeek products and any third-party services that use DeepSeek models from their information technology systems. That is a significant complication: it is not only the DeepSeek app or API that must go, but anything vendors sold into government that quietly runs on DeepSeek underneath. Second, it bars agencies from acquiring DeepSeek products or services in the future.

A parallel July memo from Homeland Security Secretary Markwayne Mullin — made public the same day — shows this did not come out of nowhere. In that memo, agencies were told to monitor traffic flowing in and out of government websites for signs of DeepSeek-related risk. Monday's order is the enforcement follow-through: the monitoring phase identified the threat picture, and the Federal Register notice is the response.

The stated reasons: three threat vectors

The DHS notice laid out why it considers DeepSeek dangerous, in language unusually blunt for a federal procurement document. It said DeepSeek models pose "multiple threat vectors": their use "may result in the transfer of sensitive data to the PRC if used by US government employees or vendors"; the models could "produce actionable intelligence on US persons and/or government agencies"; and they "may contain malware or other security vulnerabilities."

Strip away the jargon and three concerns emerge. The first is data flow: if a government employee pastes work documents into a DeepSeek-powered tool, or if a vendor's system routes queries through Chinese infrastructure, sensitive American data could end up in the People's Republic of China. The second is analysis: an AI model that processes enough government interactions could, in theory, be used to build up intelligence pictures of American officials, operations or infrastructure. The third is the code itself: the model or its surrounding software could contain hidden capabilities — malware by another name — that no user would notice until it was too late. As with all such claims, the government has presented these as risk assessments rather than published evidence, and DeepSeek has historically denied that it poses such risks.

Why it matters: AI is now infrastructure — and a battleground

This order matters beyond one company because it crystallises a shift in how the US government treats artificial intelligence. AI models are no longer just consumer products; they are infrastructure that sits inside government workflows, touching decisions, documents and data. Once that is true, every foreign model becomes a potential vector, and Washington is drawing a line: American government systems will run on AI the government trusts, and the government does not trust DeepSeek.

The timing also matters. The order landed just days after DeepSeek released a new model, V4.1-Flash-1017, billed as "smarter, faster, more efficient." The announcement was a reminder that the company is still advancing — and still competing head-to-head with American labs. In the same week that Anthropic committed 5% of its shares to a public benefit fund ahead of a blockbuster IPO and Google pushed forward with Gemini 4 Argon, the American AI ecosystem is in a full commercial sprint. The DHS ban is the security mirror of that commercial sprint: Washington is trying to make sure the race is not lost inside its own agencies.

There is also a geopolitical layer. US–China relations on technology have been defined this year by a series of calibrated confrontations and negotiations — from trade talks to technology controls. The broader trend includes China moving to restrict exports of next-generation solar technology, and both sides contesting the rules of the tech competition. AI models sit at the very centre of that contest, because whoever leads in AI may lead in everything the models enable, from cyber operations to economic productivity.

What the FCC vote could add

The DHS order may not be the last American government action against Chinese AI this month. The Federal Communications Commission is scheduled to vote on 29 October on a proposal that would bar all Chinese laboratories from testing electronics — smartphones, cameras, computers — for use in the United States. That measure is aimed at the physical device supply chain rather than software, but its logic is the same: keep Chinese-built technology, whether hardware or models, out of systems Washington deems sensitive. Together, the DHS order and the coming FCC vote show a government treating Chinese technology as a systemic risk to be managed at every layer — software, models and hardware alike.

Key facts

  • What: The US Department of Homeland Security ordered civilian federal agencies to remove DeepSeek products and third-party services using DeepSeek models from government IT systems, and banned future acquisition of DeepSeek products/services by those agencies.
  • When: Notice published Monday, 19 October 2026, in the Federal Register; a July 2026 memo from DHS Secretary Markwayne Mullin ordering traffic monitoring for DeepSeek risks was made public the same day.
  • Why (per DHS): "Multiple threat vectors" — possible transfer of sensitive data to the PRC through government employee or vendor use; potential to "produce actionable intelligence on US persons and/or government agencies"; possible embedded "malware or other security vulnerabilities."
  • Scope: Applies to civilian agencies (not, for now, the military — the Navy already issued its own ban). Includes third-party vendor products built on DeepSeek models, not just DeepSeek's own tools.
  • Prior actions: Commerce Department and US Navy previously banned DeepSeek; states including New York, Tennessee and Virginia took similar steps.
  • Next on the calendar: FCC vote on 29 October on barring all Chinese labs from testing electronics for US use.
  • Context: Order arrived days after DeepSeek released model V4.1-Flash-1017, billed as "smarter, faster, more efficient."

What happens next

Three things are worth watching. First, compliance: identifying every third-party product that quietly uses a DeepSeek model is a genuinely hard task for large agencies and their vendors, and the removal process will take weeks or months, not days. Expect follow-up guidance and possibly extensions for systems that are difficult to untangle. Second, the legal and commercial fallout: vendors whose products are built on DeepSeek will need to swap in alternative models — a costly migration that will benefit American and allied AI providers. Cyberattacks hitting financial institutions this month have already sharpened the security conversation, and the DeepSeek ban adds AI supply-chain risk to that picture. Third, the escalation path: the FCC's 29 October vote on Chinese labs testing electronics could be followed by further moves — against other Chinese AI firms, or against specific product categories — as part of the broader US–China technology confrontation, which also includes the trade framework talks in Kuala Lumpur covered in US and China Reach Trade Framework.

The bigger question: can you ban a model at all?

There is an irony that follows every AI ban, and it is worth stating plainly. A procurement ban is effective inside government networks, where IT departments can enforce what software is installed. It is far less effective in the wild: DeepSeek's models have open-weight releases that anyone can download and run locally, and researchers and developers outside government will keep using them regardless of what the Federal Register says. What Washington is really buying is hygiene — ensuring its own data does not flow into systems it cannot audit — not the disappearance of DeepSeek from the world. For American households and businesses, the practical takeaway is subtler: the US government has now put its strongest official stamp on the claim that a Chinese AI model poses national security risks, and that claim will shape insurance, procurement and regulation for years. Whether those risks were as concrete as the language suggests is a question only the classified evidence can answer.