Chinese-speaking crews use AI agents (CyberStrikeAI, 44 agents) to break into exposed legacy web systems, VPNs and WebLogic in Japan

Trend Micro reconstructed three attack infrastructures from attacker open directories, all run by Simplified-Chinese-speaking operators focused on Japan (published in Japanese 2026-10-23, in English 2026-10-27). The activity is separate from the ARTEX attacks. Case A (October 2026): the open-source CyberStrikeAI framework ran 44 AI agents that automated recon, exploitation, C2, persistence and exfiltration against at least 19 organisations (16 in Japan, also Taiwan and Hong Kong). The targets were legacy web systems at smaller businesses and local public bodies, and traffic per target was throttled to avoid notice. The deepest intrusion started at an unauthenticated, internet-exposed admin interface. Case B (Aug-Sep 2026): a human-led crew used an LLM for planning and tools. It scanned 6,000+ Japanese IPs, fully compromised legacy web systems (FTP, DB, web shell, SSH, backdoors restored after cleanup), stole data from a GIS server with default credentials, and tried to hijack exposed AI/GPU inference services. Case C (Jan-Oct 2026): a crew targeted exposed VPN and edge devices (Cisco SD-WAN, Citrix NetScaler, F5 BIG-IP) and 1,300+ Oracle WebLogic servers. It logged into Yamaha RTX1210 routers using credentials left in obtained config files. No new vulnerability was involved. Organisations with internet-exposed legacy admin panels, default or leftover credentials, or edge devices, especially in Japan, South Korea and Taiwan, are at risk. Victim names and IOCs were not published.

Summary written from the sources below. Check the originals before acting.