Useable Fully Homomorphic Encryption

Abstract

Fully Homomorphic Encryption (FHE) enables arbitrary computations to be performed over encrypted data, eliminating the need to decrypt the data and expose it to potential risk while in use. FHE promises to significantly broaden the range of applications that can be secured with End-to-End encryption. In the last decade, FHE has undergone several breakthroughs and advancements that led to a leap in performance improvements, enabling a variety of applications and a first wave of real-world deployments. However, the complexity of developing an efficient FHE application still hinders deploying FHE in practice and at scale. FHE presents unique challenges in development and deployment, which are moving to the foreground as FHE is transitioning from theory to practice. Secure computation techniques such as FHE are inherently interleaved with application logic, as they introduce both theoretical (e.g., data independence) and practical (e.g., cost model) paradigm changes. Programs need to be translated to the unique programming model of FHE, taking into account the security, expressiveness, and performance characteristics of the underlying schemes. Beyond performance, which has been the focus of the community for most of the last decade, and the challenges of development, practical deployments introduce further challenges that have so far received scant attention. Specifically, we need to carefully question to what extent the traditional threat models used in FHE (e.g., semi-honest servers and IND-CPA security) are sufficient for real-world deployments. In order to widen the set of scenarios in which FHE can be deployed effectively, we must define stronger notions of security and develop new constructions to achieve them efficiently. This dissertation presents three contributions toward useable FHE: First, we study, categorize, and distill the challenges of FHE development, identifying key characteristics that define FHE’s unique programming paradigm. Second, we introduce HECO, a Fully Homomorphic Encryption compiler that translates high-level programs to optimized FHE implementations, enabling non-experts to develop secure and efficient FHE applications. Finally, we present verifiable FHE, a new notion of maliciously secure integrity-preserving FHE that addresses the challenges arising from the mismatch between the traditional threat models used in FHE and real-world deployment scenarios.