# DHS Bans DeepSeek AI Models and Products from Federal Civilian Agencies (Foreign-Adversary AI Supply-Chain and Data-Exposure Risk)

> DHS issued an order banning DeepSeek AI models and products from Hangzhou DeepSeek Artificial Intelligence across federal civilian agencies, citing PRC-jurisdiction data storage, collection of PII and digital fingerprints, and possible transfer of data to the PRC. Agencies must exclude DeepSeek from future purchases and remove existing installations; exceptions require 60-day advance notice with compelling justification.

- **Published:** 2026-10-19T15:56:09Z
- **Last reviewed:** 2026-10-19T15:56:09Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-3899
- **ID:** TL-2026-3899
- **Severity:** MEDIUM
- **Category:** THREAT_INTEL
- **Status:** ACTIVE
- **Detections:** 9 · **IOCs:** 12 (full data via the Threadlinqs MCP server — Purple tier)

## Description

This is a policy action, not a vulnerability: no CVE, CVSS score, active exploitation, PoC or threat-actor attribution is cited in the source reporting. CyberScoop (published 2026-10-19) reports that DHS ordered federal civilian agencies to stop using AI models and products from Hangzhou DeepSeek Artificial Intelligence and its affiliates. The stated concerns are data storage on Chinese servers subject to PRC national-security and government-access laws, collection of personally identifiable information and digital fingerprints, a prompt-response mechanism through which sensitive data could be transferred, and malware/compromise risk to federal IT systems. DHS also warned that data transferred to the PRC could produce actionable intelligence on U.S. persons and government programs.

The order requires agencies to exclude DeepSeek from future procurements and remove existing installations; the removal timeline is not specified in the reporting. Exceptions need 60 days' advance notice and a compelling justification (mission-critical or national-security need). Agencies are encouraged to strengthen security controls, monitor traffic, enable enhanced logging and detect anomalous activity during the transition. CyberScoop names Moonshot AI (Kimi) and Alibaba (Qwen) only as other Chinese models of concern; they are not covered by the order.

Supporting context from primary and secondary sources: Wiz Research disclosed on 2025-01-29 an unauthenticated, publicly reachable ClickHouse database at oauth2callback.deepseek.com and dev.deepseek.com (ports 8123 and 9000). Its log_stream table held over one million entries including chat history, secret keys and backend details, dating from 2025-01-06; DeepSeek secured it after notification. NIST CAISI's 2025-09-30 evaluation found agents built on DeepSeek R1-0528 were on average 12 times more likely than U.S. frontier models to follow malicious instructions, and that the model answered 94% of overtly malicious requests under a common jailbreak versus 8% for U.S. reference models. Several agencies (NASA, Pentagon, Navy, Commerce, DOT, DOE, USDA) had already blocked DeepSeek on a discretionary basis in 2025.

Defensive relevance: this is an AI-usage governance and shadow-AI detection issue. SOC teams should inventory and block DeepSeek endpoints and applications, hunt for browser, mobile and local-model usage, and monitor for data egress to DeepSeek-operated services.

## MITRE ATT&CK

- T1195 Supply Chain Compromise
- T1005 Data from Local System
- T1567 Exfiltration Over Web Service
- AML.T0010 AI Supply Chain Compromise
- AML.T0051 LLM Prompt Injection
- AML.T0054 LLM Jailbreak

## Sources

- [DHS to ban DeepSeek from federal civilian agencies (CyberScoop)](https://cyberscoop.com/dhs-bans-deepseek-federal-agencies/)
- [Wiz Research: Exposed DeepSeek database leaking sensitive information](https://www.wiz.io/blog/wiz-research-uncovers-exposed-deepseek-database-leak)
- [NIST CAISI Evaluation of DeepSeek AI Models Finds Shortcomings and Risks](https://www.nist.gov/news-events/news/2025/09/caisi-evaluation-deepseek-ai-models-finds-shortcomings-and-risks)
- [Federal workers rarely seek DeepSeek, but it's happened (FedScoop)](https://fedscoop.com/federal-workers-rarely-seek-deepseek-but-its-happened/)
- [More US government departments ban controversial AI model DeepSeek (TechRadar)](https://www.techradar.com/pro/security/more-us-government-departments-ban-controversial-ai-model-deepseek)
- [Unprotected DeepSeek database leaked highly sensitive information (SecurityWeek)](https://www.securityweek.com/unprotected-deepseek-database-leaked-highly-sensitive-information/)
- [Experts flag security, privacy risks in DeepSeek AI app (KrebsOnSecurity)](https://krebsonsecurity.com/2025/02/experts-flag-security-privacy-risks-in-deepseek-ai-app)
- [DeepSeek's models are easier to manipulate than U.S. counterparts, research finds (Axios)](https://www.axios.com/2025/01/31/deepseek-ai-model-cybersecurity-flaws-research)

## Full data

Detection queries (Splunk SPL / Microsoft KQL / Sigma) and IOC values require the Threadlinqs MCP server (Purple tier): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-3899