Chinese-Speaking Operator Uses AI Agents for Multi-Sector Intrusions

In early 2026, threat researchers identified a campaign by a suspected Chinese-speaking operator utilizing specialized AI orchestration via the 'SecFlow' framework. This framework employed models like Claude, Qwen, and DeepSeek to coordinate reconnaissance, exploitation, and reporting. The campaign targeted a wide array of victims including the Fengtai District government, Indonesia’s Ministry of Foreign Affairs, and various educational and industrial systems across Taiwan, Vietnam, and China.

The technical workflow involved an advanced infrastructure cluster using open directories for AI orchestration, payload distribution, and fake MySQL services. The operator combined eight known CVEs with custom 'GLUTTON' steganographic webshells that hide malicious bytecode within PNG image pixels to bypass security inspections. Post-exploitation focused heavily on credential theft, including LSASS memory dumps and registry hive exports, and the deployment of a private Go-based implant named 'SecBox'.

This campaign demonstrates the maturation of AI-driven offensive operations, where automated agents manage complex attack chains. The impact includes the confirmed compromise of government administrative records, patient health documents, and student profile information from university systems, highlighting a significant risk to data confidentiality and integrity across the Asian region.