Wiz Research details how misconfigurations in LiteLLM, an open-source AI gateway, can lead to cloud compromise. The report highlights that nearly one in ten internet-facing LiteLLM servers accepted the default ‘sk-1234’ admin key, allowing unauthenticated access. This vulnerability, combined with unauthenticated MCP sessions and custom code guardrails, exposes cloud AI infrastructure to root-level remote code execution and IAM theft.