DeepSeek Ban: DHS's June 26 Order Bars US Agencies From Buying AI

The US Department of Homeland Security is barring federal civilian agencies from buying information technology products and services from DeepSeek's parent company, in a Federal Register notice published Tuesday. The exclusion covers all IT products from Hangzhou DeepSeek Artificial Intelligence and its affiliates, and it requires agencies to remove any DeepSeek versions already installed on their systems.

The timing is the story. The notice follows the earlier US accusations, first reported in February, that DeepSeek's makers systematically distilled output from US frontier labs to gain an edge, and it lands inside a broader Washington push to keep the most capable open-weight models out of foreign state reach, even as Beijing casts the curbs as economic coercion dressed up as security.

Related reading: Nvidia Expects Chip Sales to Double: The Filings Back It.

The notice, published in the Federal Register on Tuesday as document 2026-21447 (91 FR 66346), rests on two authorities: the Federal Acquisition Supply Chain Act and a review by the Federal Acquisition Security Council. That council, made up of agency security and acquisition officials, is the body the law created to exclude risky technology vendors from government purchasing. The notice includes a copy of the letter from Secretary Markwayne Mullin ordering the exclusion on June 26, first reported by CyberScoop, which means the policy was in force behind the scenes for months before the notice appeared.

What agencies must actually do is the practical core. They have to exclude DeepSeek products from all future procurements and strip out any versions already in place. Exceptions are available but costly to obtain: an agency must submit its request 60 days in advance and provide a compelling justification, such as an inability to perform a mission-critical function or a separate national security consideration, according to CyberScoop. Agencies are also encouraged to harden their networks while removal is under way, by monitoring traffic, adding data exfiltration controls, and flagging anomalous activity.

Related reading: A 60-Day Clock on the Federal Definition of 'Super Intelligence' Starts Today. Nvidia, Meta and the Labs Are Watching.

Mullin's stated rationale tracks the arguments US frontier firms have made for more than a year. He wrote that DeepSeek's models present multiple threat vectors to civilian agency IT systems, a claim first reported by CyberScoop, and he pointed to where DeepSeek stores data. Models hosted directly by DeepSeek collect personally identifiable information and trace users' online activity, he said, and data held in China sits within reach of national security laws that give the government broad latitude to request it. He further noted that every major AI model uses a prompt-and-response process that can transfer sensitive data, and that, combined with publicly available information, could produce actionable intelligence on US persons and programs.

DeepSeek is the only Chinese model named in the order, but it is not the only one under scrutiny. Moonshot AI, maker of the Kimi assistant, and Alibaba, maker of the Qwen family, have also drawn national security attention and have been accused by US firms of illegally distilling their models, according to CyberScoop. That makes this notice a single, named enforcement action inside a broader posture toward Chinese open-weight models, rather than an isolated ban.

Related reading: HPE Books a $1.2 Billion AI Order From Vultr Without an Nvidia Rack, as the Buildout's Value Moves to the Network.

For markets, the direct ticker impact is close to zero because DeepSeek is private and has no US listing. The significance is structural. The federal government is now treating Chinese open-weight AI not just as a chip-export concern at the compute layer but as a procurement and supply-chain risk at the software layer. That widens the line Washington is drawing: restrict the silicon, then restrict the models that run on it.

The compute layer is where the money sits, and it is supply-constrained. This desk reported earlier that ByteDance, Alibaba and Tencent ordered more than 3 million of Nvidia's export-compliant cards worth over $35 billion, orders Nvidia has warned it may fill only in part. A software-layer ban now sits on top of that supply squeeze, pushing the open-weights-versus-closed debate from a developer argument into a government purchasing decision that favors vendors able to show where models and user data are stored and who can legally reach them.

China responded within hours. The foreign ministry said the ban disregards a consensus reached between Xi Jinping and Donald Trump, according to Reuters, and Beijing has defended DeepSeek against what it calls an overstretching of the concept of national security. No retaliation specific to the notice had been announced as of this writing.

What is left open is the enforcement clock. Exception requests start their 60-day advance window when agencies file them, agency removal timelines have not been published, and the more consequential question is whether the Federal Acquisition Security Council extends the same treatment to Moonshot or Alibaba, which would widen the exclusion beyond a single vendor. For readers of the AI supply chain, the dated facts to track are those removal timelines, the first exception decisions, and any China countermeasure, all of which carry market consequences for the vendors who depend on federal and state AI procurement.

This content is for informational purposes only and does not constitute financial advice. Past performance is not indicative of future results. Consult a qualified financial advisor before making investment decisions.