
Amazon Uncovers AI Agent Campaign Targeting LiteLLM, Dify and Other Apps for API Keys
Amazon Threat Intelligence has detailed an ongoing campaign in which autonomous AI agents systematically scan for exposed instances of LiteLLM, Dify, One API and similar platforms to extract stored AI provider keys. The agents, assessed with moderate confidence to be operated by China-linked groups, use captured keys to power further reconnaissance and exploitation while maintaining a shared pool of working credentials. Records obtained via MadPot decoy servers reveal a detailed playbook covering default credentials, authentication bypasses via the isCliRequest header, SSRF to cloud metadata services, and known flaws such as CVE-2025-3248 in Langflow and CVE-2026-28797 in RAGFlow. Parallel activity observed by Cloudflare showed ARTEX agents abusing static CI/CD tokens to access customer environments and create persistence via Worker scripts. The same framework has also been used in attempts to bypass cryptocurrency exchange identity checks with deepfakes. Defenders are advised to eliminate default keys, enforce token expiry and IP restrictions, and monitor for the published ARTEX indicators.
The Hacker NewsNov 16, 2026Translated from English