FBI Warns Artex AI Agents Are Automating Intrusions and Credential Theft
Emerging Technology Risk · 17 sources · Updated 45m ago · First seen Oct 3, 2026

The FBI warned that cybercriminals and nation-state actors are using Artex, an open-source agentic penetration-testing framework, to automate attacks against government agencies and critical sectors worldwide. Artex translates natural-language objectives into commands executed by AI agents, enabling parallel operations with limited supervision; some intrusions reportedly reached privileged access in under three hours. Reported activity includes attacks on South Korean and Japanese financial institutions and a Chinese state-backed supply-chain attack against ConnectWise. Although its original GitHub repository was removed, Artex remains available through forks, archives, localized variants and a commercial version. Censys identified more than 1,600 publicly exposed consoles.
Amazon Threat Intelligence’s MadPot sensors captured AI-agent requests containing campaign histories between October 3 and November 5, 2026. Amazon attributed the campaigns to Artex users and disparate Chinese-nexus threat groups with moderate confidence, but could not independently verify activity outside its sensors or agent-reported outcomes. A prominent campaign targeted internet-exposed AI applications to steal provider credentials, reportedly reusing stolen keys to fund further credential hunting. Another targeted two cryptocurrency exchanges with forged identity documents and real-time deepfakes, but every recorded deepfake liveness-verification attempt failed. The findings underscore the need to patch exposed applications, restrict management interfaces, protect provider credentials, and monitor AI-service usage and spending. Although high-volume agent activity can trigger defensive sensors, its speed can sharply compress response windows.


