Threat · Breaking

FBI warns AI hacking framework ARTEX is running intrusions at machine speed; Amazon ties it to Chinese-nexus groups stealing AI model keys

The FBI issued a FLASH alert (FLASH-20261113-001) on Nov 13 warning that ARTEX, an open-source agentic penetration-testing framework, is being used by criminals and state actors to automate intrusions from reconnaissance to data theft, sometimes reaching privileged access in under three hours. The same day, Amazon said disparate Chinese-nexus groups are using ARTEX agents to steal AI model API keys at scale, with stolen keys paying for the AI that hunts for more.

Nov 13, 2026 · 2 min read

What happened. The FBI released FLASH-20261113-001 on Nov 13 with TTPs and IOCs for ARTEX, an open-source, LLM-driven multi-agent pen-test platform written in Go. The FBI says a variety of actors have used it since at least August 2026 against financial services, government, technology, defense, semiconductor, energy, healthcare, manufacturing, telecom and legal organizations. It names two campaigns: attacks on South Korean financial institutions that took personal data on about 67,000 people, and the ConnectWise Cloud Backup supply chain compromise that gave access to more than 2,250 downstream tenants. The original developer pulled the GitHub repo in October, but forks, archived copies and a commercial ARTEX PRO keep it in circulation; the FBI cites over 1,600 publicly exposed ARTEX consoles as of early November.

Amazon's piece. Amazon Threat Intelligence (post by CISO CJ Moses) says its MadPot decoy sensors, set up to look like popular LLM inference APIs, caught ARTEX agents from Oct 3 to Nov 5 sending requests that carried the agents' own campaign notes. Amazon assesses with moderate confidence these are disparate Chinese-nexus groups. The biggest campaign stole AI model API keys and cloud credentials, with stolen keys feeding the framework's model gateway to power more key hunting. Targets were exposed AI gateways and app builders such as LiteLLM, One API, New API, Dify, Langflow, Open WebUI and Ollama, because one weak admin panel can hold many providers' keys. A second campaign tried to beat crypto exchanges' identity checks (KYC) with real-time deepfakes.

Why this matters to an enterprise. ARTEX doesn't need a new zero-day. The FBI says it goes after the systems nobody watches closely, like partner and supplier portals, mobile backends, employee apps and legacy systems, then moves fast with valid credentials. Practical steps from the alert: put rate limits and volume alerts on high-impact business APIs, enforce server-side authorization on customer-record endpoints, and watch AI API billing and token use for spikes. If you run an internal AI gateway (a proxy that holds your OpenAI, Anthropic or other model keys), take its admin page off the internet and rotate the keys it holds. ConnectWise customers should review every Microsoft 365 tenant where the affected apps were ever authorized, back to at least Oct 1.

Background: Check Point Research and Microsoft have also published on ARTEX abuse this month; the FBI alert is marked TLP:CLEAR and can be shared freely.

Sources