AivexaNewsSearch
AI news for builders and product teamsChecked every hour

US orders federal agencies to remove China’s DeepSeek AI from their systems

Collected Oct 20, 2026

The US government has ordered federal civilian agencies to halt purchases of products and services from Chinese AI startup DeepSeek and to strip the company's technology from government IT systems. The directive, attributed to the Department of Homeland Security, cites national security concerns tied to DeepSeek's operations in China. It covers not only DeepSeek's own products but also third-party products and services built using DeepSeek technology, and it applies both to new purchases and to systems already in place. Agencies can request exceptions.

The stated rationale centers on data handling. According to the order, sensitive information belonging to US government users could potentially be accessible to Chinese authorities under Chinese law. That framing is the crux of the restriction: the concern is not a specific demonstrated breach but the legal environment in which a China-based provider operates. For agencies, the practical effect is an inventory-and-removal exercise that extends past the DeepSeek brand itself, since any vendor embedding DeepSeek models or APIs in a government-facing product falls inside the scope.

This is not the first US restriction on DeepSeek. A law passed in December 2025 targeted its use in intelligence community national security systems. The new order widens the perimeter to civilian agencies, which is a much broader footprint across departments that buy commercial software and cloud services.

China pushed back. Foreign Ministry spokesperson Mao Ning said Tuesday that Beijing opposes what she described as overstretching the concept of national security and politicizing technology and trade issues, and added that China will resolutely safeguard the legitimate rights and interests of Chinese enterprises, according to an official briefing.

The order lands amid intensifying AI competition between the US and China. DeepSeek has challenged US rivals including OpenAI and Meta with low-cost open-weight models, a positioning that made it a reference point for teams weighing price against capability. Separately, Bloomberg reported last week that China is considering export restrictions on 23 AI-related technologies, including chip designs and certain advanced AI models, citing security and misuse risks.

DeepSeek has drawn scrutiny from governments and regulators globally since its R1 reasoning model shook the industry in January 2025 and triggered a sharp sell-off in Nvidia shares. The US has also accused DeepSeek and five other Chinese AI companies, among them Alibaba and Moonshot AI, of using large-scale model distillation to extract information from American AI systems, per a September report. OpenAI and Anthropic have each raised separate concerns about alleged distillation. OpenAI told US lawmakers it had seen suspected attempts by DeepSeek to distill its models, while Anthropic said it identified unauthorized attempts by DeepSeek and other Chinese labs to collect outputs from Claude. Distillation, in this context, means training one model on another model's outputs to approximate its behavior at lower cost; it sits in a gray zone between legitimate benchmarking and unauthorized extraction, depending on terms of service and scale.

For developers and product teams, the immediate question is supply chain visibility. A model provider is now also a compliance variable, and the order's extension to third-party services means a government customer may ask which underlying models a product calls, where inference runs and who can access logs. This suggests vendors serving public-sector buyers will increasingly need model-level disclosure and swap-ready architecture so a component can be replaced without a rewrite. A likely trade-off is cost: DeepSeek's appeal has been low pricing on capable open-weight models, and substituting a US-hosted alternative may raise inference bills and add latency or integration work.

Why it matters: Federal civilian agencies and their software suppliers must now audit and remove DeepSeek-derived components, and exception requests create a paper trail for anything that stays. The broader signal, as an inference, is that model provenance is becoming a procurement requirement rather than a technical footnote, which raises diligence burdens for any team embedding third-party models in regulated or government-adjacent products.

Read at TechCrunch · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

The U.S. has ordered federal civilian agencies to remove DeepSeek products and third-party services built using its technology, citing national security concerns.