Amazon Uncovers AI Agent Campaign Targeting LiteLLM, Dify, and Other Apps for API Keys
Our summary
Amazon's threat intelligence unit, using decoy servers from its MadPot network, captured logs showing AI-powered attack agents probing internet-facing AI platforms like LiteLLM, One API, Dify, FastGPT, Langflow, RAGFlow, Open WebUI, LobeChat and Ollama to harvest AI provider API keys. With moderate confidence, Amazon believes the activity runs on an open-source penetration-testing framework called ARTEX, likely operated by separate China-linked groups, which chains planner and worker agents driven by an LLM to automate reconnaissance and exploitation. The agents exploited known weaknesses including Langflow's CVE-2025-3248 (missing authentication, fixed in 1.3.0), RAGFlow's CVE-2026-28797 (server-side template injection, patched in 0.25.0), an authentication bypass via the isCliRequest header in One API-family tools, default/hard-coded admin credentials like LiteLLM's master key, and SSRF against cloud metadata endpoints. Separately, Cloudflare reported ARTEX compromising a CI/CD API token lacking IP restrictions, triggering hundreds of database dump attempts in an hour, while the FBI warned that multiple threat actors have abused ARTEX since August across sectors worldwide, including a South Korean breach exposing data on roughly 67,000 people.
The Hacker News publishes this story on its own site; we link to it rather than reprint it.
Worried this affects your company?
Discuss a security assessment of your internet-facing systems. Scope agreed before testing.