Chinese Hacker Commands DeepSeek via Telegram to Launch Autonomous Attacks

A Chinese-speaking threat actor leveraged the DeepSeek large language model inside the Hermes Agent framework to autonomously discover and exploit vulnerabilities in internet-exposed workflow tools (Langflow, n8n, Marimo) with minimal human involvement after a single Telegram instruction. This represents a significant escalation in attacker capability: AI agents can now perform reconnaissance, target selection, and exploitation at machine speed and scale, reducing the window between vulnerability disclosure and active exploitation. The attack surface consisted of over 460 publicly reachable systems, many of which likely lacked timely patching or adequate exposure controls. This matters because agentic AI dramatically lowers the skill barrier for sustained, autonomous campaigns, meaning even moderately resourced adversaries can execute complex attack chains. Organizations that have not minimized their internet-facing footprint and maintained aggressive patch cycles are disproportionately at risk.