The DeepSeek order is also a dependency-inventory problem

The Department of Homeland Security published notice of an order excluding DeepSeek-developed IT products and services, including those of affiliates, from federal civilian agency procurement. It also orders removal of DeepSeek and affiliate products, plus third-party products and services developed using DeepSeek implementations, from civilian agency IT.

That last phrase is the operational challenge. An agency can remove a known chatbot and still have a supplier tool whose model dependency is not visible in its software inventory. The notice does not itself provide a list of every affected third-party product. It also does not announce a ban on private-sector or consumer use across the United States.

1. 1

   Find direct use. Identify installed applications, sanctioned AI services, API integrations and accounts that use DeepSeek or its affiliates in covered environments.

2. 2

   Find indirect use. Ask vendors which models and model providers sit behind their AI features, whether implementations can change, and whether the product uses hosted inference or local model weights.

3. 3

   Record the decision. For each product, document the supplier answer, the deployment and data flow, the order’s applicability, the replacement or removal owner, and any evidence needed for closure.

This order shows a gap many organisations already have: they know which application they bought but not which model it calls or where prompts travel. Even outside the federal requirement, a useful inventory should distinguish the model developer, hosting provider, deployment mode and data sent to the service. Those facts support a real risk decision; a generic “secure AI” label does not.