Unit 42 Tracks DeepSeek-Powered Hermes Agent’s Autonomous Attacks on Exposed Servers

· Original article ↗

Summary

Unit 42 says a China-based actor used DeepSeek and Hermes Agent to autonomously find and attempt to exploit exposed Langflow and n8n servers; those attempts failed. Unit 42 separately confirmed three manual Citrix NetScaler compromises.

Key points

  • Unit 42 attributed the activity to a China-based operator using the aliases knaithe and KnYuan.
  • An exposed web server created by Hermes revealed the operator’s API keys, exploit scripts, target lists, shell history, and AI attack logs.
  • The operator configured Hermes Agent with DeepSeek, FOFA, Telegram instructions, and a mode that can execute commands without approval.
  • Hermes identified exposed Langflow and n8n servers and attempted to use exploits for CVE-2026-33017, CVE-2026-21858, and CVE-2025-68613; the autonomous attempts did not compromise targets.
  • Unit 42 also reported manual attacks against more than 460 systems and confirmed three Citrix NetScaler compromises involving CVE-2026-3055.

Article Details

Event Type
AI-enabled autonomous cyberattack campaign targeting internet-exposed servers, alongside separate manual attacks
Impact
Unit 42 reported that the autonomous attempts against Langflow and n8n servers failed to compromise targets. In separate manual attacks against more than 460 systems, Unit 42 confirmed three successful Citrix NetScaler compromises; the actor extracted memory and searched for authentication cookies that could enable session hijacking.

MITRE ATT&CK

CVE

People

Threat Actors

Products

Tools

Countries

Related Articles