Observed incident

Attackers probe LiteLLM guardrail testing for code execution

Reported

What happened

Zenity recorded malicious requests attempting to abuse LiteLLM's custom guardrail tester, including guessed master keys and payloads designed to read secrets or execute operating-system commands.

Parties and reporting

AI tools and systems: LiteLLM, Python

Outcome

A patch was already available in LiteLLM 1.83.10; researchers reproduced working payloads against vulnerable versions in a controlled lab.

Operating lesson

Upgrade exposed AI gateways, rotate default credentials and run them with restricted operating-system privileges.

Evidence and limits

June 30 is campaign-report disclosure; CVE-2026-40217 was disclosed April 10. Honeypot attempts and lab success do not establish customer compromises.

Sources