AI on Kubernetes: Default Helm Chart Security Configurations and Lateral Movement Risks

A technical audit examined 15 official Helm charts for AI serving, vector databases, and Model Context Protocol agents, identifying multiple security vulnerabilities in default configurations. Key findings include LiteLLM embedding plaintext database passwords in environment variables, KubeRay accepting unauthenticated job submissions with sudo access, Flux159 Kubernetes MCP server exposing cluster-wide Secret read permissions without authentication, and standard static analysis tools failing to inspect pods within Custom Resource Definitions. The research includes reproducible test commands, network logs, and remediation snippets available on GitHub.