Three LiteLLM flaws let low-privilege users gain admin access and run code, exposing AI keys, secrets, prompts, and responses ...