Confidential Virtual Machines (CVMs) protect guest memory and execution state from an untrusted hypervisor, but conventional observability mechanisms may expose sensitive information. This thesis investigates how useful operational visibility can be retained without undermining the confidentiality guarantees of CVMs.