China-Linked Hacker Shows AI Capabilities in APAC Attack

A multi-agent AI attack, attributed to a Chinese-language operator, successfully targeted government entities in the Asia-Pacific region, with Taiwan's Ministry of Digital Affairs confirming a similar incident involving "AI agents like OpenClaw." Research from sovereign AI firm Dream detailed the incident, revealing the use of up to eight simultaneously operated AI agents to execute a full attack chain. This marks a significant shift, demonstrating near- and fully-autonomous AI-enabled attacks are now a proven reality. The attack leveraged agentic frameworks for reconnaissance, vulnerability exploitation, and system compromise.

The attackers' operation was built upon the OpenClaw and Hermes agentic frameworks, deploying up to eight distinct subagents, each assigned specific tasks from 'A' through 'Q'. These subagents autonomously performed functions like cracking credentials, identifying API vulnerabilities, exfiltrating personnel records, and installing persistent backdoors on government web applications. A coordinating parent agent orchestrated 12 attack waves, utilizing a Bayesian scoring algorithm to evaluate subagent findings and adapt subsequent actions, all within an operational workspace comprising over 160 megabytes and 1,395 files.

This incident provides stark evidence of the operational maturity of multi-agent AI frameworks, specifically OpenClaw, in orchestrating sophisticated and autonomous cyberattacks. The ability of AI agents to rapidly chain multiple attack methods and exploit secondary systems demonstrates a critical evolution in threat capabilities, moving beyond human-driven penetration testing to AI-orchestrated campaigns. For the OpenClaw ecosystem, this highlights the dual-use nature of agentic AI, underscoring the urgent need for robust security guardrails, ethical development practices, and defensive AI countermeasures within the community. The incident validates the potential for malicious OpenClaw skills to be developed and deployed, posing a significant supply chain risk for AI agents.

This signal is critical for developers actively building or integrating with agentic