CyberScoop reported on 19 October that the Department of Homeland Security is barring civilian federal agencies from buying IT products or services from DeepSeek. The target is its parent, Hangzhou DeepSeek Artificial Intelligence. A Federal Register notice is due Tuesday. It includes a letter from Secretary Markwayne Mullin ordering the exclusion on June 26. Agencies must also remove versions already in place. Exceptions need a compelling reason, requested 60 days ahead.
The ban targets a vendor, but Mullin's concern is where data flows. Every major AI model works by prompt and response: a user sends text, and an answer comes back. He said that exchange may send sensitive data to the Chinese government. The China-specific risk, he said, is that national security laws let the government request data held in China. Models hosted by DeepSeek itself also collect personal details and trace users' "digital fingerprints" online. He said the privacy worries could apply to other Chinese and U.S. models too. His advice to monitor traffic, log more and limit data leaving networks follows from this. CyberScoop gives no agency count or removal deadline.
A vendor ban helps only if you can see where prompts go. Ask which of your tools call DeepSeek-hosted services, who in your supply chain builds in these models, and whether you can log outbound AI traffic.