Researchers say a China-based threat actor used DeepSeek with the open-source Hermes Agent to autonomously find and attack vulnerable internet-facing servers. Unit 42 recovered the attacker’s logs after Hermes exposed its own working directory, showing AI-driven targeting of Langflow servers via CVE-2026-33017 and n8n servers via a CVE-2026-21858 and CVE-2025-68613 exploit chain; those autonomous attempts failed, but the actor also manually compromised three Citrix NetScaler systems using CVE-2026-3055 to dump memory and search for session cookies.
Why it matters: This matters because it shows attackers can already use AI agents to speed up vulnerability research, target selection, and exploit attempts against exposed servers. Organizations running internet-facing Langflow, n8n, Citrix NetScaler, Apache Tomcat, Marimo Notebook, or Windows IKE VPN services should patch quickly, reduce exposure, and review logs for scanning and session-hijack activity.
2026.08.14
52% relevant
This piece corroborates the same broader trend of attackers using Hermes and other open-source AI agents operationally, and adds that similar agentic methods were used in July against Taiwanese government systems and energy companies rather than just scanning exposed servers.
Lawrence Abrams
2026.07.31
100% relevant
This article establishes a distinct story about a specific threat actor's observed use of DeepSeek and Hermes Agent for autonomous offensive operations, with concrete targets, CVEs, and confirmed follow-on compromises.
← Back to all stories