Google's LLMJacking research shows why security teams need visibility into cloud identities, model API keys and unexpected AI-service consumption.