Encryption-In-Use for Cloud-Native Workloads: Integrating Homomorphic Encryption and Confidential Computing in Kubernetes
Abstract
While cryptographic measures can be applied to the data which is either stationary or operational, there are certain risks to sensitive information during processing unless the computer environment has been secured. In cloud-native systems, this leaves a practical choice. Hardware-backed Trusted Execution Environments (TEEs) preserve conventional application semantics with modest steady-state overhead, while Fully Homomorphic Encryption (FHE) keeps selected data encrypted during computation at the cost of substantial algorithmic and operational constraints. This paper presents a Kubernetes-oriented architecture that combines Confidential Containers, remote attestation, policy-governed secret release, and selective FHE. The contribution is not the general idea of splitting computation between a TEE and FHE, which has been explored in prior hybrid systems, but an operationalization of that pattern for Kubernetes with explicit trust boundaries, attestation workflows, and data-field partitioning. A threat-model-driven selection framework is provided for standard Kubernetes, confidential virtual machines, workload-level CoCo, pure FHE, and a hybrid TEE-FHE design. Quantitative platform observations are drawn only from cited prior studies, while a 1,000-record transaction case study is analyzed at the operator level to show how policy-driven field partitioning can eliminate encrypted comparisons and masking multiplications. The same architectural pattern is relevant to privacy-preserving Generative AI serving, where encrypted linear operators can be delegated while non-linear and control-heavy operations remain inside an attested TEE.