The FBI tells that the Artex AI tool speeds up attacks
The FBI tells that some Artex attacks went from first target to privileged access in less than three hours. Copies of the tool stay available.
On Friday the FBI told the public about Artex, an open-source AI tool that makes network tests. The FBI tells that criminals and groups that work for governments use it to break into government agencies and many important sectors worldwide. It connects a large language model (LLM, an AI text model) to attack tasks and does them at machine speed.
How it works
A user writes a goal. AI agents (software that acts without help) change it into machine commands. Artex works with Claude, GPT, DeepSeek, GLM, Grok and Kimi. The FBI tells that DeepSeek Flash models are used most. They cost less and show no guardrails when the user calls the work a penetration test.
The numbers
The FBI tells that some attacks went from first target to privileged access in less than three hours. Censys, a cyber firm, found more than 1,600 open Artex consoles. Amazon tells that one group used keys that it took from others to pay for AI use that then looked for more keys.
What is not known
The FBI tells that the total number is probably more, because systems behind firewalls are not seen. The developer, named as Li Puhua, denied responsibility and removed the code from GitHub. The FBI tells that forks, copies and a commercial version keep it available.
Insight: Removing the code did not remove the threat
The developer removed Artex in October. The FBI tells that forks and a commercial version continue. Check Point Research found 1,608 live consoles. This reads as a tool that nobody owns. The control points are the guardrails of AI models and weak side doors. Leaders cannot rely on a takedown.
Sources
Posted
