The FBI warns that criminals use the ARTEX AI tool in attacks
Some attacks went from the first target to admin access in less than 3 hours, says the FBI.
The FBI published a Flash notice on November 13 about ARTEX, a public AI tool for security tests of computer systems. Criminal groups use it against organizations around the world since August 2026, and some attacks went to privileged access in less than 3 hours. One attack got personal data of about 67,000 persons from South Korean financial institutions.
How it works
ARTEX does not have an AI model. It uses external large language models (LLMs, AI text models).
The DeepSeek Flash family is used more than other models. The FBI gives the reasons: low cost and no guardrails that it can see when the work is a security test.
The numbers
The FBI gives 2 examples:
- South Korean financial institutions: personal data of about 67,000 persons.
- ConnectWise Cloud Backup: a supplier attack gave access to more than 2250 customer accounts.
What is not known
The creator removed the code from GitHub in October. The creator says the tool is only for tests with permission and that the creator had no part in the attacks.
Copies of the tool continue to be used, with ScopeWeaver and a commercial version, ARTEX PRO.
Insight: The tool code is public, but the AI model is a gate
The creator removed the code, but Censys found 1,465 open ARTEX control panels in November. The FBI says the tool uses external models. The most used models have no guardrails when the work is a security test. This reads as a gate: copies of the code stay in use, and the model providers are the one point that can stop it. Monitor if they add checks.
Sources
- FBI Highlights Growing Abuse of AI-Driven Pentesting Framework ARTEXinfosecurity-magazine.com
Posted