AI-driven infrastructures demand observability pipelines that are both intelligent and hardware-trusted. Conventional Kubernetes observability tools such as Fluent Bit, OpenTelemetry, and Loki provide deep insights but risk exposing sensitive telemetry data during processing. This work presents a confidential observability framework using confidential-compute-enabled Kubernetes nodes. By leveraging hardware-based isolation, telemetry remains encrypted and is processed only within attested workloads, ensuring end-to-end confidentiality. Integrating CNCF observability components with secure hardware advances trustworthy AI operations, enabling privacy-preserving analytics, regulatory compliance, and resilient monitoring essential for future intelligent ecosystems across finance, healthcare, and autonomous digital systems.