Morning Edition · Tuesday, October 20, 2026Published at 2:12 AM EDT · New York
The order is based on a 2018 supply-chain security law. Because it also covers US-hosted deployments and products built on DeepSeek's models, it extends beyond DeepSeek's own app to the underlying open-weight models.

The Department of Homeland Security (DHS) has ordered federal civilian agencies to remove products made by DeepSeek, the Chinese AI lab, from their IT systems and to stop buying them, The Hill reported. Homeland Security Secretary Markwayne Mullin signed the order on July 24, and it appeared on the Federal Register website on Monday.
The scope reaches past DeepSeek's own app and application programming interface. The order covers DeepSeek and DeepSeek-affiliate products, and also "third-party products and services developed using DeepSeek implementations," according to BankInfoSecurity. The AI Post channel noted in its summary of the order that hosting the model in the US does not exempt a deployment. IBTimes read the clause as covering tools built on DeepSeek's open-weight models.
DHS issued the order under the Federal Acquisition Supply Chain Security Act of 2018 (FASCSA), after reviewing a recommendation from the Federal Acquisition Security Council (FASC). The department says DeepSeek's models present "multiple threat vectors." It argues that the Chinese government has "extensive authority" to access private infrastructure, that agency use could send data on federal employees and contractors to Beijing, and that the models themselves might contain malware or security flaws (Korea JoongAng Daily).
That list contains two separate arguments. The data-exfiltration argument applies to DeepSeek's hosted service, where prompts travel to servers DeepSeek controls. It does not apply to weights served on a US-controlled graphics processing unit (GPU). The coverage of derivative products therefore depends on the second argument, that the weights themselves could carry hidden behavior. DHS asserts that risk, but the published reporting cites no demonstrated backdoor in DeepSeek's weights. The reporting also does not explain how the order treats fine-tuned versions, models distilled from DeepSeek's outputs, or models that reuse DeepSeek's published architecture.
Beijing has not issued a direct response to this report. When DHS signed the order in July, Liu Pengyu, a spokesperson for China's embassy in Washington, called congressional allegations of technology theft "nothing but slander and smears." China's Foreign Ministry spokesperson Guo Jiakun said China opposed "the broadening of the concept of national security," CyberScoop reported.
The order governs what runs inside federal agencies. It does not stop attackers from using DeepSeek's models as tools elsewhere. In a separate offensive campaign reported this week, a Chinese-speaking actor reached DeepSeek, Kimi and Anthropic's Claude Opus 4.6 through the OpenRouter aggregator, Computing reported. That campaign used models from both the United States and China.
Part of a tracked trend
Chinese Open-Weight Models Emerge as the Non-US AI Stack
As Washington restricts foreign access to US frontier models, governments and enterprises cut off from American AI increasingly standardize on downloadable Chinese open-weight models, splitting the world into competing AI supply blocs rather than a single frontier.
Start a discussion in Townsquare.
More from this edition
If the order reaches derivative products, US-origin open-weight suppliers such as Meta and Google gain share among federal software vendors. Contractors that built DeepSeek weights into their products must pay for audits and replacement, and DeepSeek itself loses little because federal civilian agencies were not its customers.
Several outlets confirm that Homeland Security Secretary Markwayne Mullin signed Department of Homeland Security (DHS) Exclusion and Removal Order 2026-01 on July 24 and that it covers products "developed using DeepSeek implementations." Still, no published text says whether self-hosted, fine-tuned or distilled weights count, DHS has shown no backdoor in the weights, and the order excludes national security systems. Beijing has objected to an "overgeneralization of national security concepts" since February 2025, and this desk found no new Chinese statement on the order's publication.
An open-source-intelligence read of how likely this story is true with its real nuance, not a judgment of any outlet. It assesses the claim, weighing independent and adversarial reporting. How we label confidence.
What this means
The order works through federal procurement. The parties exposed are software vendors and integrators that sell to civilian agencies and embed DeepSeek weights, fine-tunes or distillations in their products, along with cloud model catalogs that serve government accounts. DeepSeek itself loses almost no revenue, because US federal buyers were never its customers. The gainers are US-origin open-weight suppliers such as Meta and Google, whose models become the default choice for vendors selling to the government. The consensus view treats the order as largely symbolic because it was signed back in July. This desk reads the derivative-products clause as the substantive part, because it makes a model's origin a compliance property that buyers will have to audit. Two precedents point the same way. DHS ordered Kaspersky software off federal systems in 2017, and the Commerce Department banned Kaspersky sales to US consumers in 2024. Section 889 of the 2019 defense authorization act first barred Huawei and ZTE equipment from federal purchases, then extended that bar to contractors that used the equipment. The likelier outcome is that within months, federal contractors begin requiring model-provenance attestations (which base model, which distillation data) in software bills of materials, and that pressure builds to apply the same treatment to other Chinese labs. This reading depends on the FASC continuing to issue orders under FASCSA. DHS guidance that narrows "DeepSeek implementations" to DeepSeek's own hosted services would prove it wrong.
What to watch
Observations to monitor, not financial advice.
Source: Polylog editors
Comments
0No comments yet.