US bans DeepSeek AI from federal civilian agencies over security risks

The US Department of Homeland Security banned DeepSeek AI and its affiliated products from all federal civilian agency IT systems and procurements on October 16, 2026. The order applies to agencies including the Environmental Protection Agency and the Federal Communications Commission, while excluding national security systems and secure compartmented information systems operated by the Department of Defense and the intelligence community. Federal agencies can seek written exemptions within 60 days if they demonstrate compelling reasons, such as mission-critical operations or national security, though they must monitor network traffic and strengthen controls until removal.

The ban stems from concerns over ‘multiple threat vectors,’ including potential malware, security vulnerabilities, and unauthorized data transfers to China. The Department of Homeland Security highlighted risks tied to DeepSeek’s server locations in China, where authorities can compel companies to hand over user data under local law without notification. DeepSeek collects sensitive user data such as keystroke patterns, IP addresses, and device identifiers, further raising national security alarms.

DeepSeek, founded in 2023 by Liang Wenfeng – who also co-founded the Chinese hedge fund High-Flyer – gained global attention in early 2025 for releasing large language models that rivaled US competitors like ChatGPT and Google Gemini at significantly lower costs. The company’s flagship model, R1, was particularly noted for its reasoning capabilities and cost-effectiveness, while its V4 model saw a permanent 75% price reduction in 2026, further boosting its adoption. DeepSeek releases its models as open-weight, allowing third parties to download and use the underlying model weights, which has embedded its technology into numerous third-party products and services globally, including those offered by Snowflake, Vercel, and OpenRouter.

China’s Foreign Ministry spokesperson Mao Ning condemned the ban on October 20, 2026, calling it an ‘overstretching of the concept of national security’ and accusing the US of politicizing technology issues. Beijing vowed to ‘resolutely safeguard the legitimate rights and interests of Chinese companies,’ including DeepSeek, in response to the restrictions. The move follows a pattern of state-level bans on DeepSeek, with Texas prohibiting its use on government devices in January 2025, followed by New York, Virginia, and Tennessee in February 2025. Legislation introduced in 2025, including the No DeepSeek on Government Devices Act (S.765) and the Protection Against Foreign Adversarial Artificial Intelligence Act, sought to formalize these restrictions at the federal level.

The ban coincides with broader US-China tensions in AI governance. Last week, China proposed new export restrictions on advanced AI models, escalating reciprocal measures after US-led bans on China’s access to advanced AI chips and manufacturing equipment. While the US and China have made recent progress in AI governance collaboration – such as a proposed notification system for national security incidents – the two nations remain deeply divided on key issues, including data sovereignty and technology transfer. Model distillation, a technique used by Chinese AI developers to extract capabilities from larger US AI models to train smaller ones, has intensified competition and distrust between the two nations.

US AI companies, including Anthropic and OpenAI, have accused Chinese developers of using this method to reproduce capabilities from their systems. The upcoming meeting between US President Donald Trump and Chinese President Xi Jinping in mid-November 2026 at the APEC leaders’ summit in Shenzhen may address these tensions.