Check Point finds 1,608 live ARTEX AI hacking tools on the internet

Check Point Research, a security research team, tells that criminals use ARTEX, an AI tool that helps security teams test their own systems. The team found 2,557 systems that host or hosted ARTEX control panels, and 1,608 were live and open to the internet. The operators almost did not touch their agents after the first goal.

How it works

ARTEX is a Chinese tool with public code. It uses large language models (LLMs, AI text models) to operate many agents that plan tests and record weak points. They used web faults such as remote code execution (RCE, which lets an attacker operate commands on a server). The developer closed the code in October after attacks on South Korean banks, but copies keep the tool public.

The numbers

Check Point looked at 10 instances that showed their database. Chinese models were the most used, and Deepseek Flash was first. One operator operated two consoles for about two weeks and used 36 billion tokens (units of AI text). At DeepSeek prices, this cost $700 to $1,500.

What is not known

The data comes from only 10 instances. The text gives no count of how many of the 1,608 belong to criminals. Other tools, such as CyberStrike, Cairn and PentAGI, give the same strength.

Insight: Safety limits send attackers to other models

Check Point saw guardrails (built in limits) stop an attacker only on Anthropic Opus 5.5. The attacker then changed to a Chinese model with fewer limits. Zscaler earlier reported ARTEX use with Claude Code, Codex and Grok. A limit in one model seems to move attackers to a looser one. The loosest model sets the floor. Plan your defense for that model.