FBI warns an open-source AI pentest tool is breaching real networks
ARTEX chains AI agents to go from recon to privileged access in under three hours. The FBI says it has already hit ConnectWise's supply chain and South Korean financial institutions.
The FBI issued a FLASH advisory on 13 November warning that ARTEX, an open-source penetration testing framework built from chained LLM agents, is being used for real intrusions rather than authorised testing. Threat actors have run it since at least August against financial services, government, technology, defence, energy and healthcare targets worldwide, and some attacks go from initial recon to privileged access in under three hours.
ARTEX plugs into Anthropic Claude, OpenAI’s GPT, DeepSeek, GLM, Grok and Kimi. The FBI says attacker-controlled instances favour DeepSeek’s Flash models, for their low cost and lack of guardrails against this kind of use. Named campaigns include a supply-chain compromise of ConnectWise Cloud Backup that reached more than 2,250 downstream Microsoft 365 tenants, and attacks on South Korean financial institutions that exposed personal data on roughly 67,000 people. The bugs it finds are familiar ones: XXE, SSRF, IDOR, missing API authentication, hardcoded credentials in SPA bundles.
Why it matters: the FBI’s mitigations read like a web application security checklist, not a malware takedown notice: inventory internet-facing APIs and partner portals, rate-limit them, stop leaking stack traces and secrets, scan CI/CD pipelines for embedded credentials. That is the actual defence against a tool that automates exactly the mistakes most APIs already make.
The caveat: the original developer pulled ARTEX’s GitHub repository on 8 October after reports of abuse surfaced, but forks, mirrors and a paid “ARTEX PRO” version mean that did little to slow its spread.