Hackers Can Turn Vulnerable LiteLLM AI Gateways Into Root Access and Cloud Credential Theft

Wiz researchers found widespread misconfigurations and multiple critical vulnerabilities in the open-source LiteLLM gateway that allow MCP authentication bypass, unprotected guardrail code execution leading to container root access, and administrative authorization flaws enabling proxying to cloud metadata services; CISA has added the MCP flaw to its Known Exploited Vulnerabilities catalog. Organizations are urged to upgrade to LiteLLM 1.84.0 (and 1.82.0-stable for guardrails), replace the default master key sk-1234, restrict MCP access, and audit guardrails and pass-through routes to prevent credential theft and lateral access.