A China-linked group used AI agents on more than 200 organizations

CrowdStrike says a group with the name VAPOR PANDA used ARTEX from October 17, 2026. ARTEX is a tool that lets AI agents do attacks. The group used it against more than 200 organizations in 14 or more countries. Targets included foreign ministries and telecom providers. CrowdStrike confirmed break-ins at nine organizations. It says the group attacks many targets at the same time, not one.

How it works

ARTEX has one primary agent. Other agents select the steps and do the work, and all of them use one graph of facts and goals. A human operator gave instructions in Chinese at some points. These told the agents to write web shells and to find administrator passwords. The agents used DeepSeek, a Chinese language model, for most of the work.

The numbers

• CrowdStrike confirmed break-ins at nine organizations.
• The group also tried attacks on passwords on 12 or more other organizations.
• In one break-in, the agents used about 1.4 billion input tokens, the units of text that a model reads.

What is not known

CrowdStrike does not know which other language model the group used. The count of tokens is for one break-in only. The total for the campaign is more. CrowdStrike gave no test of the agents against manual work. It says other sources report different ARTEX attacks as Storm-0940.

Insight: One open tool serves thieves and spies

In October 2026, attackers used ARTEX against South Korean banks, and Zscaler says the instructions point to a goal of money. CrowdStrike links ARTEX to a China-linked group that told its agents to find computers with no CrowdStrike sensor. Open agent tools seem to make attacks easy for all types of attackers, and leaders must count computers with no sensor.