TLDR Tech / InfoSec
Attackers Exploit Exposed AI Endpoints for Offensive Operations
Attackers are exploiting misconfigured Ollama and LiteLLM instances by targeting exposed inference endpoints that frequently lack authentication or rely on vulnerable placeholder keys. Three observed groups used autonomous penetration testing frameworks and a Codex-based agent for probing, weaponizing tools, and performing web reverse-engineering via hijacked backends. Defenders should ensure no internet exposure, enforce robust authentication, inspect request bodies, and monitor abused ports and IP addresses.
Read the full story