The AI agent story has a dark twin. ARTEX is an autonomous penetration-testing agent — connect it to an LLM and it picks targets, plans intrusion routes and runs hacking tools by itself. In October, investigators in Tokyo found it on the servers behind a breach wave.
On October 19, GMO Cybersecurity by IERAE published what it found at victims' request: over September 18 to October 18, six overseas servers were used against Japanese organizations — and all six carried traces of ARTEX, and no other AI agent. The infrastructure targeted about 20 Japanese companies, nine in the same industry (insurance, finance and logistics concentrated), and eight had disclosed breaches. GMO also found six further AI-based attack infrastructures and handed its findings to law enforcement.
The number to be careful with
You may see "around 80 companies targeted" — that figure comes from Wall Street Journal reporting on the wider campaign and does not appear in GMO's own post. GMO's documented scope is ~20. Both can be honest at once, but they are different kinds of claim — one forensic, one reported — and the difference matters. GMO also named no actor: the evidence points to infrastructure, not people.
The Korea chapter
ARTEX first surfaced against South Korean financial institutions in late September (CrowdStrike reported on it October 7, including data-exfiltration cases). Its developer halted public releases on October 8 — but the tool had been public since July, and copies already out do not expire. A pentesting agent published for defenders is, line for line, an attack agent for anyone else.
Why it worked
Japan was already mid-wave: 119 disclosed web-system compromises by October 6, 2026, versus 84 in all of 2025. The conditions are not unique to Japan — the language barrier fell, public web apps and APIs were left to age, and shared suppliers turn one breach into dozens. The attackers automated first.
Full investigation with the defence checklist and FAQ: ARTEX guide on ToolPilot Pro
Top comments (0)