The Decoder reported on 21 October that the US Department of Homeland Security told civilian federal agencies to remove all Deepseek products from their IT systems. The order also covers other vendors' software built on Deepseek's open-source models. It bars future purchases too. Washington fears Beijing could reach sensitive data, the outlet said. China's Foreign Ministry says the US stretches national security too far. Its Commerce Ministry wants the ban lifted and has promised countermeasures.
The Decoder says China does not approve ChatGPT, Claude or Gemini. Since 2023, its Cyberspace Administration has had to review AI services. The sources do not say what China's countermeasures will be. They do not say how agencies should find Deepseek models inside other vendors' products. They give no exact date for the US order.
Our reading, not the outlet's: the main exposure may be a model hidden inside another vendor's product, not the Deepseek app. An open-source model is one anyone can download and build into their own tool, so the label may not name it. Buyers can ask each vendor three things. Which model does the product call? Does it run on the vendor's servers or on yours? Can the vendor list every model inside it, like a parts list?