US federal agencies ordered to remove Chinese AI service DeepSeek

Who is affected

The US Department of Homeland Security (DHS) has ordered all federal civilian agencies to stop using technology from DeepSeek, a Chinese artificial intelligence (AI) lab. The order also covers any DeepSeek-affiliated products and third-party services developed using DeepSeek implementations. This means not only DeepSeek's own software, but also any tools or services that might use DeepSeek's AI models behind the scenes. The Register reports that complying will be difficult, as agencies may not always know which products rely on DeepSeek, especially when using AI gateways like OpenRouter or Vercel AI Gateway that can route requests to DeepSeek models without users realising it. The DHS has not yet clarified the full scope of the ban.The Register↗ (opens in a new tab)

The response

The DHS order follows earlier moves by other US government bodies. The Pentagon's Defense Information Systems Agency blocked DeepSeek in January 2025. The US Navy, NASA, and the House of Representatives have also banned the service. Some US states have taken similar action. Lawmakers previously proposed a bill to ban DeepSeek on government devices, citing national security concerns, but it did not pass. Despite this, agencies have acted independently to restrict DeepSeek's use.

What experts say

Katie Moussouris, CEO of Luta Security, told The Register that banning DeepSeek's own app and API is practical, but the order's broad language makes compliance almost impossible. She points out that the order could technically cover any software developed with help from DeepSeek, even if only used during code writing or testing. There is currently no way for vendors to prove their products never used DeepSeek, since software bills of materials do not track which AI tools developers used. Moussouris also notes that the main risks cited, data flowing to DeepSeek's servers in China and possible vulnerabilities, are not unique to DeepSeek, and US-hosted versions of the models would avoid most of those concerns.

How the flaw works

Concerns about DeepSeek include the possibility that its AI models could capture sensitive data and send it to servers in China. Past reports on DeepSeek's mobile apps for iOS and Android found issues such as plaintext data leaks, connections to Chinese tech firm servers, and lack of SSL pinning (a security feature that helps prevent man-in-the-middle attacks). The US National Institute of Standards and Technology (NIST) found DeepSeek more vulnerable to prompt injection (where attackers trick AI models into doing things they should not) than some US models, and easier to jailbreak (bypass safety controls). However, these are problems seen in many AI models, not just DeepSeek.

What you should do

For government agencies and contractors, the main action is to review their use of AI services and check whether any systems or software rely on DeepSeek or its models, directly or indirectly. This includes checking with vendors and being cautious with AI gateways that may route requests to DeepSeek. For the wider security community, the case highlights the challenge of tracking the origins and dependencies of AI models in complex software supply chains.