Trend Micro details three AI assisted attacks targeting Japan

In autumn 2026, Trend Micro researchers studied three attack infrastructures to understand how hackers use AI against Japanese targets. The National Cybersecurity Office in Japan issued a warning on 9 October 2026, and researchers note that AI-assisted activity varied across the cases. Case A saw AI agents manage the entire attack workflow, including reconnaissance, vulnerability checks, exploitation, data theft and reporting, using CyberStrikeAI—an open‑source penetration-testing platform built around AI agents.

The platform split tasks among multiple AI agents, with one record mentioning a credential-search agent in a system running 44 agents; it could generate Japanese forms and fake login pages. Researchers identified at least 19 target organisations, with at least 16 in Japan, and observed multi‑target, low-traffic activity that obscured hits when inspecting logs from a single victim.

Often there was no active session or web shell, and the clearest intrusion started via an administration page on an old web system, not a newly discovered vulnerability.

Case B involved a human operator who used a large language model to assist with planning and tool development. The plan, allegedly presented as authorized security testing, targeted AI and GPU resources, including AI workflows, data processing and model inference platforms. In one campaign, the attacker gained control across several systems—FTP, a database, an administrator account, a web shell and SSH—and even restored backdoors after removal.

In another, a GIS server with default credentials exposed hundreds of thousands of land and business records, Windows credentials and email data. Case C featured AI in a more limited role, primarily aiding tool and script development rather than planning or management. All three campaigns ran in 2026, targeted exposed VPN devices and Oracle WebLogic servers, and used Simplified Chinese across infrastructures, with Japan the main target country. The researchers caution that AI-enabled activity can inflate apparent outcomes and stress the need for verification.