DeepSeek ban: US orders agencies to remove Chinese AI tools

The Department of Homeland Security directed federal civilian agencies to stop buying services and products from Chinese AI startup DeepSeek and to remove existing deployments, according to reporting by TechCrunch. The order extends beyond DeepSeek’s own product to third-party products and services built on DeepSeek technology, and agencies can request exceptions.

What the DeepSeek ban covers and why

TechCrunch reports the DHS action covers both new purchases and existing IT systems used by federal civilian agencies. The department raised concerns about how DeepSeek handles user data, warning that sensitive information from U.S. government users could potentially be accessible to Chinese authorities under Chinese law, the article says. The order therefore targets direct use of DeepSeek products and any third-party offerings that incorporate DeepSeek technology.

The move is part of wider U.S. scrutiny of certain Chinese AI labs. TechCrunch summarizes prior steps, including a law passed in December 2025 that restricted DeepSeek’s use in U.S. intelligence national security systems. The article also notes U.S. accusations that DeepSeek and five other Chinese AI companies used model distillation to extract information from American AI systems, and that companies including OpenAI and Anthropic raised related concerns.

How other governments and companies have reacted

TechCrunch reports China pushed back through its Foreign Ministry, with a spokesperson opposing what Beijing described as politicizing technology and trade issues and saying it would protect Chinese companies’ rights and interests. The article also cites Bloomberg reporting that China is considering export curbs on 23 AI-related technologies.

TechCrunch places this action in a wider context: DeepSeek’s R1 model drew international attention in January 2025, and the company has faced scrutiny from regulators and other governments since then.

What this means for agencies, vendors, and partners

For federal civilian agencies the immediate effect is a halt to new purchases of DeepSeek-based products and a requirement to remove existing deployments unless an exception is approved, TechCrunch reports. For vendors and integrators, the order means suppliers should expect government customers to request proof about what models and components are in their products, and to face potential procurement limits where DeepSeek technology is present.

The order also highlights broader risks that drove the decision, namely concerns about data access under foreign law and the use of techniques like model distillation that other companies have complained about, as reported by TechCrunch.

Practical steps for affected organizations

Based on the facts reported by TechCrunch, consider these actions:

• Inventory any products, services, or components that use DeepSeek technology, including third-party offerings.
• Pause new purchases that reference DeepSeek until you have clear guidance from procurement or counsel.
• Ask suppliers to document the origins of the AI models and whether DeepSeek technology was used in training or distillation.
• Limit sensitive data exposure to tools whose data handling and jurisdictional controls you cannot verify.
• Maintain or tighten access controls and multi-factor authentication for systems that previously integrated DeepSeek tools.
• If you are a federal agency, follow DHS procedures to request an exception only when necessary and document the security rationale.
• Monitor official DHS guidance and any follow-up notices for implementation details.

These steps are general good practice that flow from the concerns TechCrunch describes about data access and national security, and from the practical effect of an order that covers third-party products.

What to watch next

TechCrunch notes the decision follows earlier U.S. restrictions and industry complaints about model distillation. Expect additional government guidance on compliance and exception processes. Keep an eye on possible policy moves from China, such as the Bloomberg-reported export consideration for AI-related technologies, which could affect suppliers and cross-border model availability.

If your organization is evaluating AI vendors and you need a plain-English plan for how to assess vendor risk and model provenance, consider a simple roadmap to prioritize safety and procurement controls. If helpful, our studio offers a basic AI roadmap to guide those decisions: Get a plain-English AI roadmap.

This briefing uses reporting from TechCrunch.