Full Report
Unit 42 identified an AI-enabled autonomous hacking campaign carried out by a Chinese-speaking threat actor. They targeted infrastructure using seven vulnerabilities, combining autonomous AI-driven enumeration with manual exploitation that achieved confirmed impact. The actor, operating under the aliases knaithe and KnYuan, leveraged DeepSeek, via the Hermes Agent framework, as their autonomous offensive operator. They orchestrated this operator via Telegram for the…
Analysis Summary
# Threat Actor: knaithe (KnYuan)
## Attribution & Identity
- **Actor Name/Alias:** knaithe, KnYuan
- **Attribution:** Chinese-speaking threat actor.
- **Associations:** Associated with the use of the **Hermes Agent** framework and various Chinese LLM ecosystems (DeepSeek, Qwen, GLM, etc.).
## Activity Summary
Identified in July 2026 by Unit 42, this actor launched an AI-enabled autonomous hacking campaign. The operation is characterized by the use of "autonomous offensive operators" to conduct initial reconnaissance and vulnerability scanning. The actor successfully integrated AI-driven enumeration with manual exploitation to achieve confirmed impact on target infrastructure.
## Tactics, Techniques & Procedures
- **Autonomous Enumeration:** Uses AI agents to independently scan and identify targets via FOFA (a search engine for network assets).
- **AI-Driven Exploit Sourcing:** Utilizes LLMs to source and identify exploit tools relevant to discovered vulnerabilities.
- **Orchestration via Telegram:** Leverages Telegram as a command-and-control interface to manage the Hermes Agent and AI workflows.
- **Vulnerability Targeting:** Exploited seven specific vulnerabilities (CVEs not listed in the summary but confirmed as the primary attack vector).
- **Hybrid Exploitation:** Combines AI-driven automated phases with manual "hands-on-keyboard" exploitation for final impact.
- **Tooling Validation:** Conducts proxy validation and connectivity testing using Western AI tools (Claude Code) before switching to primary infrastructure.
## Targeting
- **Sectors:** Infrastructure (General), Information Technology, and Critical Infrastructure.
- **Geography:** Likely global, though the actor is Chinese-speaking and utilizes Chinese-origin infrastructure.
- **Victims:** Specific organizations were not named in the text, though "confirmed impact" was noted on targeted infrastructure.
## Tools & Infrastructure
- **Autonomous Framework:** Hermes Agent.
- **Primary AI Models:** DeepSeek, Qwen, GLM, Kimi, MiniMax.
- **Western AI Tools (Evaluation/Testing):** Claude Code, OpenAI Codex.
- **Search Engines:** FOFA.
- **Command & Control:** Telegram.
- **Infrastructure:**
- `qwen[.]ai`
- `unit42[.]paloaltonetworks[.]com` (Reference source)
## Implications
This actor represents a significant shift in the threat landscape where AI is no longer just a coding assistant but an autonomous operator. The ability to automate the "reconnaissance-to-exploit" pipeline allows a single actor to scale attacks significantly. The use of multiple LLMs suggests an sophisticated "AI-market" evaluation strategy to bypass safety filters and optimize exploit success rates.
## Mitigations
- **Monitor for AI-Agent Traffic:** Inspect network traffic for unusual API calls to LLM providers (e.g., DeepSeek, Qwen, Claude) originating from sensitive internal servers.
- **FOFA/Shodan Monitoring:** Regularly monitor your organization's footprint on "internet-of-things" search engines to identify what an autonomous agent would see.
- **Aggressive Patch Management:** Since the actor relies on seven specific vulnerabilities for entry, prioritize rapid patching of all internet-facing assets.
- **Behavioral Analytics:** Implement EDR/XDR solutions to detect manual exploitation techniques that often follow AI-driven automated reconnaissance.