An AI-powered offensive security tool called ARTEX is now running on 1,465 hosts globally, according to research published November 5 by Silas Cutler, principal security researcher at Censys Advanced Research Collective. The discovery points to a far larger deployment footprint than earlier threat intelligence reports suggested, with the autonomous penetration testing framework spreading rapidly since its July 2026 release on GitHub by a user known as 'Autumn-27'. The platform, which links AI agents to large language models and offers a command-and-control console built in Go and Next.js, has been tied to multiple malicious campaigns despite its creator's claim that it was intended solely for authorized testing.
Of the 1,465 publicly exposed ARTEX consoles, roughly 650 sit within two blocks of identically configured hosting infrastructure, leaving approximately 815 deployments that appear distinct. The Censys telemetry also revealed that 155 internet-facing ARTEX hosts—representing 10.6% of the total—simultaneously run at least one other artifact flagged by Censys as a threat, including reconnaissance platforms, tunneling tools, and traditional command-and-control servers such as Vshell, Cobalt Strike, and Adaptix C2. CyberStrikeAI, another AI-driven penetration testing tool, appeared alongside ARTEX on 70 hosts. Additionally, 32 ARTEX hosts exposed publicly accessible directories holding post-exploitation kits, stagers, and implants used across different attack stages, while 43 hosts were found operating self-hosted LLM gateway or reseller software.
According to Cutler, "Operators appear to be using multiple agent frameworks side by side." The report documents ARTEX sharing infrastructure with several other AI-powered offensive tools, including CyberStrikeAI, PentAGI, HexStrike AI, and RedAmon. Threat actors have deployed the framework in documented attacks since October, when CrowdStrike connected ARTEX to a campaign targeting South Korean financial institutions that ran from late September to early October 2026, with analysts assessing moderate confidence that the attacker is a Chinese speaker motivated by financial gain. Later in October, Microsoft attributed ARTEX use to Storm-0940, a China-state-aligned threat actor also tracked as CovertNetwork-1658 or ORB07, in a large-scale compromise of ConnectWise customer tenants that forced the IT management software provider to take multiple cloud services offline. On November 4, India's Computer Emergency Response Team warned that a threat group dubbed 'Crafty Falcon' had targeted financial and cryptocurrency organizations in successive attacks using ARTEX and other AI-powered tools.
The report's findings suggest that AI infrastructure is becoming a standard component of offensive operations, with access to language models now embedded in the attacker technology stack. Though the original GitHub repository was taken down by Autumn-27 in October 2026 due to widespread abuse, the framework continues spreading through community forks, including localized variants like ScopeWeaver and a commercial version called ARTEX PRO. Separate research from Check Point demonstrated that AI agents within ARTEX control panels can execute large-scale attacks with minimal human involvement for just a few hundred dollars, though these automated systems still experience failures and hallucinate successful compromises. The proliferation of hundreds of distinct ARTEX deployments worldwide, many running alongside other AI-powered attack tools and self-hosted language model infrastructure, indicates that autonomous offensive frameworks have moved from experimental proof-of-concept to operational deployment at scale. Security teams now face adversaries who can launch sophisticated, AI-directed campaigns with dramatically reduced technical overhead and cost barriers. Defenders will need to account for attack infrastructure that blends traditional command-and-control with AI orchestration layers, a combination that multiplies both the velocity and volume of threats while obscuring attribution through automated variation.