Censys counts 1,465 exposed ARTEX consoles: attack agents have the same misconfiguration problem as everyone else
Censys calls it agentic asbestos. As of November 4, the scanning firm counted 1,465 internet-facing hosts running an exposed console for ARTEX, the open-source autonomous pentest framework released in July and later tied to the South Korean finance intrusions and Storm-0940's ConnectWise campaign. TechCrunch says these could be reached without a password.
Keep the numbers straight, because coverage already differs. About 650 of those hosts sit in two blocks of identically configured hosting, which suggests a provisioning template rather than hundreds of operators. Take those out and you get roughly 815 distinct-looking deployments, which is TechCrunch's "more than 800". Censys says global deployment is likely higher, since the console doesn't need to be public at all. And 155 hosts, or 10.6%, also carry at least one other threat detection, which is the more useful signal for separating operators from testers. Censys itself says some of these are probably authorized testers.
The exposed data is the awkward part. Censys says the instance from CrowdStrike's Korean bank report had been publicly exposed for months, and its file names show work well beyond those banks, including payment and gaming services. So the people targeted by these agents may have their details sitting in a console anyone could open. Researchers, and other attackers, get to read the same thing.
There's also a closure worth noting. When we wrote in October that going closed-source wouldn't fix anything, the argument was that copies were already out. Censys confirms it. The repository is gone and the author said there would be no more releases, but archived copies remain, and TechCrunch counts dozens on code-sharing sites. Removing the source removed updates, not the capability.
The lesson runs both ways. Autonomous tooling has become easy to stand up, and the operators standing it up are skipping the basics, just as defenders and startups have with their own agent dashboards. If you run any agent console yourself, whether for red-teaming or anything else, this is a reason to check whether it can be reached from the internet, whether it requires a login, and what past run data sits behind it. No patch is involved. A console that holds tools, credentials and run history is a high-value target no matter whose side it's on. Censys: https://censys.com/blog/artex-agentic-asbestos/ TechCrunch: https://techcrunch.com/2026/11/06/hundreds-of-artex-ai-hacking-tools-found-publicly-exposed-to-the-web/ Our October piece: https://clauday.com/article/ce9a836f-f86e-4106-8c16-fa14608170a4
← Back to all articles
Keep the numbers straight, because coverage already differs. About 650 of those hosts sit in two blocks of identically configured hosting, which suggests a provisioning template rather than hundreds of operators. Take those out and you get roughly 815 distinct-looking deployments, which is TechCrunch's "more than 800". Censys says global deployment is likely higher, since the console doesn't need to be public at all. And 155 hosts, or 10.6%, also carry at least one other threat detection, which is the more useful signal for separating operators from testers. Censys itself says some of these are probably authorized testers.
The exposed data is the awkward part. Censys says the instance from CrowdStrike's Korean bank report had been publicly exposed for months, and its file names show work well beyond those banks, including payment and gaming services. So the people targeted by these agents may have their details sitting in a console anyone could open. Researchers, and other attackers, get to read the same thing.
There's also a closure worth noting. When we wrote in October that going closed-source wouldn't fix anything, the argument was that copies were already out. Censys confirms it. The repository is gone and the author said there would be no more releases, but archived copies remain, and TechCrunch counts dozens on code-sharing sites. Removing the source removed updates, not the capability.
The lesson runs both ways. Autonomous tooling has become easy to stand up, and the operators standing it up are skipping the basics, just as defenders and startups have with their own agent dashboards. If you run any agent console yourself, whether for red-teaming or anything else, this is a reason to check whether it can be reached from the internet, whether it requires a login, and what past run data sits behind it. No patch is involved. A console that holds tools, credentials and run history is a high-value target no matter whose side it's on. Censys: https://censys.com/blog/artex-agentic-asbestos/ TechCrunch: https://techcrunch.com/2026/11/06/hundreds-of-artex-ai-hacking-tools-found-publicly-exposed-to-the-web/ Our October piece: https://clauday.com/article/ce9a836f-f86e-4106-8c16-fa14608170a4
Comments