Artex AI Hacking Tool: 800+ Consoles Left Open Online
Censys found 1,465 passwordless Artex AI consoles online as of Nov. 4; about 815 look distinct, and 10.6 percent also run other threat tools.
Hakkı ALKANPublished 3 min read
Internet scanners at Censys reported on Nov. 5, 2026 that 1,465 hosts were running an exposed Artex AI console with no login wall as of Nov. 4. About 650 of those hosts sit in two identically configured hosting blocks, which leaves roughly 815 deployments that look distinct. TechCrunch summarized the same research on Nov. 6 as more than 800 distinct copies left open to the web.
Artex, also written ARTEX, is an open-source, LLM-driven autonomous penetration testing framework. Penetration testing means authorized probing of systems for weaknesses. The project was released on GitHub in July 2026 by Autumn-27, then moved toward closed-source after public abuse reports. The console does not need to be reachable from the public internet, so the open instances are a misconfiguration, not a product requirement.
Why Do Artex AI Host Counts Differ?
Censys matches Artex by its web-console title and Next.js assets. The raw host count is 1,465. Two providers account for about 44% of that total with cloned port layouts, so Censys treats many of those addresses as templated copies rather than unique operators. After excluding those blocks, about 815 deployments remain. That is the "800+" figure used in headlines, including our Turkish report earlier today.
Not every open console is malicious. Censys notes that authorized testers also run Artex. Still, leaving a passwordless attack console on the public internet raises the risk that outsiders can reuse the same tooling against unintended targets.
What Else Was Found on Exposed Artex Hosts?
Of the 1,465 hosts, 155 (10.6%) also run at least one other Censys-detected threat. Examples include AI attack frameworks stacked next to Artex, plus conventional command-and-control tools such as Cobalt Strike, Vshell, and Adaptix C2. Censys also saw 32 open directories on Artex hosts exposing implants, stagers, and post-exploitation kits.
In October, CrowdStrike linked Artex to a late September–early October campaign against South Korean finance targets, assessed as a Chinese-speaking financially motivated actor using DeepSeek v4.1-flash. Microsoft later tied Artex to Storm-0940 activity around a ConnectWise customer-tenant compromise. TechCrunch's Nov. 6 report notes many linked attacks still leaned on weak passwords and missing multi-factor authentication.
Our Take
Agentic hacking tools compress the time from scan to exfiltration, but the Censys findings show a simpler failure mode: operators leaving the control plane open. If you run offensive AI frameworks, keep the console off the public internet, require authentication, and watch for co-located C2 or open directories.
For earlier Artex campaign context, see our report on the Vapor Panda Artex AI attacks.
Defenders chasing the same AI arms race can also read our coverage of OpenAI's GPT-6 Cyber security model.
Related Linux remote-access risk is covered in our sshole vulnerability report.
| Item | Details |
|---|---|
| Scan date | Nov. 4, 2026 (Censys) |
| Exposed consoles | 1,465 hosts |
| Distinct-looking | ~815 after templated hosting blocks |
| Also running other threats | 155 hosts (10.6%) |
| Open directories | 32 |
| First release | July 2026 (Autumn-27 / GitHub) |