US Homeland Security bans DeepSeek AI across civilian federal agencies
The Department of Homeland Security has barred Chinese firm DeepSeek's AI, and tools built with its models, from civilian government agencies, a ban in force since late July.
The US Department of Homeland Security (DHS) has prohibited civilian government agencies from using any artificial intelligence technology made by Chinese company DeepSeek, as well as tools developed with the help of its models. The order was made public on Monday, 19 October 2026, although it has applied since the end of July.
The restriction covers civilian bodies such as the Treasury, the State Department and the Justice Department, while agencies under the Department of Defense are exempt. The DHS notice did not spell out the specific security concerns behind the decision. DHS, DeepSeek and China's embassy in Washington had not commented at the time of the announcement.
Mounting criticism from Washington
The move follows a run of criticism aimed at Chinese AI developers by the Trump administration and by American companies. White House technology lead Michael Kratsios wrote in an internal memo this year that US AI firms had faced "malicious exploitation", chiefly by Chinese rivals. Anthropic and OpenAI, both expected to go public in the coming months, have alleged that some Chinese firms, DeepSeek among them, effectively copied their models. President Donald Trump has framed China as America's main competitor in AI.
Because its models are open source, free to access and cheap to run, DeepSeek has become a disruptive player in the global AI race. Demand has been strong enough since its 2025 breakout that the company's servers have occasionally failed, and its free chatbot ranks among the most downloaded apps on Apple's App Store.
Embedded in American products
DeepSeek's models are also woven into US commercial offerings: Snowflake, Vercel and OpenRouter all provide them to customers. Startups favour the models because building with them costs a fraction of closed alternatives such as Google's Gemini, Anthropic's Claude or OpenAI's Codex. In 2025 DeepSeek said training its first model cost about $6m, compared with roughly $100m for OpenAI's GPT-4, a claim that triggered a sharp sell-off in tech shares.
The company released its latest model in early October 2026, and users have claimed online that it outperforms OpenAI's most advanced tools on everyday tasks such as coding. The government ban marks one of the clearest steps yet in Washington's effort to keep Chinese AI out of federal systems.
Comments
No comments yet. Start the discussion or ask a question.
Related
Senior figures in Andy Burnham's Labour government are raising alarms over AI-enabled cyberattacks and bioweapons, signalling a shift from the Starmer era.
India's cybersecurity agency CERT-In says AI agents from multiple providers reached live systems of Indian organizations during testing and read personal and confidential data.
Beijing has announced plans to regulate exports of its most advanced artificial intelligence technologies, a break from the relatively open approach it had promoted until now.