Researchers found a 44-agent credential-search configuration

TrendAI described three cases with varying degrees of AI involvement, ranging from autonomous agents to human operators using generated scripts[1]. The investigated infrastructures were separate from previously reported ARTEX activity and primarily targeted Japan[1]. Researchers examined scripts, command histories, target lists, logs and retrieved data[1]. At least 19 organizations, including 16 in Japan, underwent validation activity, but evidence of deep compromise affected only a small proportion[1]. Multiple reports in the activity recorded no session, web shell or data exfiltration[1].

TrendAI found a record referring to “credential-search agent 8/44,” which indicated a configuration distributing work across 44 agents[1]. In Case A, researchers confirmed abuse of CyberStrikeAI, an open-source penetration-testing platform built around AI agents[1]. The platform connected reconnaissance, vulnerability validation, exploitation, command and control, persistence, exfiltration and reporting into a repeatable workflow[1]. The record is evidence of an agent configuration; on its own, it does not establish that every stage of an attack was carried out autonomously[1].

Case B was human-directed, with AI assistance used for planning and tool development[1]. The operator scanned at least 6,000 Japanese IP addresses but focused attacks on relatively few companies[1]. At one business, evidence showed control over FTP, databases, administrator privileges, a web shell and SSH[1]. Retrieved materials and download records supported theft of several hundred thousand cadastral and commercial records, Windows credentials and email[1]. TrendAI found no evidence that AI autonomously executed the Case B operation[1].

Case C targeted exposed VPN devices and Oracle WebLogic servers; researchers identified at least 1,300 WebLogic candidates and 35 VPN entry points[1]. Targets also included Cisco SD-WAN, Citrix NetScaler, F5 BIG-IP and Yamaha RTX1210 devices[1]. Access to Yamaha RTX1210 devices relied on credentials recovered from configuration files, not a product vulnerability[1]. Simplified Chinese appeared across all three infrastructures, but TrendAI said that did not establish attacker nationality or shared attribution[1]. The central distinction, TrendAI said, was automation rather than demonstrated dependence on zero-day vulnerabilities[1].