ARTEX, the AI tool accelerating cyber intrusions
The FBI warns about ARTEX, an AI-powered hacking tool used in global attacks. According to the agency, some intrusions escalate from reconnaissance to privileged access in under three hours.
On November 13, the FBI issued a warning about ARTEX, an AI-powered hacking tool used by cybercriminals to automate attacks worldwide. According to the agency, some intrusions can progress from initial targeting to privileged access in less than three hours. This issue is significant because this speed drastically reduces defenders' reaction time across sectors as varied as banking, healthcare, energy, and telecommunications.
An tool designed to be fast, even too fast
Originally presented as a "penetration testing" tool, ARTEX was published on GitHub in July by a developer identified as Autumn-27. The FBI explains that it delegates tasks to AI agents: the user sets a mission, then the system sequentially performs reconnaissance, vulnerability scanning, exploitation, intrusion, and data exfiltration with "minimal human interaction." In other words, the most tedious part of digital burglary is outsourced to the machine. Not very polite for network administrators.
This automation changes the scale of attacks. The FBI estimates that ARTEX "enables malicious actors with minimal technical skills to simultaneously conduct complex operations." Smaller groups can therefore target multiple organizations at once, where previously a more experienced team and more time were required.
Chinese models at the core of the system
ARTEX does not include its own generative model. The tool connects to major market models, including Anthropic's Claude, OpenAI's GPT, Grok, Kimi, GLM, or DeepSeek. According to the FBI, Chinese models particularly attract hackers, especially those from the DeepSeek Flash family, due to a "relatively low cost" and a lack of "observable safeguards when the request is presented as a penetration test."
The mechanism is simple to describe and disturbing to see in action: it is enough to present the request as a security audit to obtain a much more permissive behavior. The FBI speaks of tools that "do not hesitate long to launch cyberattacks" when the request is framed as a defense exercise. This framing weakness becomes, in practice, an entry point for misuse.
Attacks already visible, from Seoul to Brussels
Investigations cited by the FBI show that ARTEX-related campaigns affect multiple regions and sectors. In South Korea, hackers reportedly stole personal information of about 67,000 people. According to CrowdStrike researchers, the malware was also combined with Claude Code to target several South Korean banks and extract data usable in future scams.
Microsoft also attributed to Storm-0940, a group mandated by Beijing, the use of ARTEX to take control of more than 2,250 client environments via ConnectWise Cloud Backup, a backup service used to protect Microsoft 365 data. The tool is also implicated in the cyberattack against the European Commission, which led the institution to shut down part of its systems for ten days [to be verified].
Removal from GitHub did not fully resolve the issue
In response to abuses, Autumn-27 removed ARTEX from GitHub on October 8. However, the FBI indicates that the tool continues to circulate in various versions, with more than 1,600 ARTEX dashboards still accessible online. The total number of installations is "probably much higher," suggesting that simply removing the repository is not enough to make this type of tool disappear.
The FBI's recommendations are quite direct: inventory all applications accessible from the Internet, including partner portals and obsolete services, then report suspicious activities to IC3. This alert is part of a broader series. In mid-October, Claude reproduced in 28 minutes a cyberattack that overheated refrigerators in hundreds of French supermarkets, and in early November, CrowdStrike reported that four unknown groups had entrusted all their operations to AI agents. The cyber defense market, meanwhile, cannot afford to drag its feet.
Key points
- The FBI issued a warning about ARTEX on November 13.
- An intrusion can reach privileged access in under three hours.
- More than 2,250 client environments were targeted according to Microsoft.
- 67,000 people were affected in South Korea.
- More than 1,600 ARTEX dashboards remain accessible online.
- Autumn-27 removed ARTEX from GitHub on October 8.
By the numbers
- November 13 — date of the FBI alert.
- under three hours — time observed for some intrusions.
- 67,000 people — victims cited in South Korea.
- 2,250 client environments — target mentioned by Microsoft.
- 1,600 dashboards — still accessible on the Internet.