Veritas - Verifiable Encodings for Secure Homomorphic Analytics

In the Veritas project, we investigated verifiable encodings for secure homomorphic analytics, i.e., cryptographic authenticators to guarantee the correctness of privacy-preserving computations performed using fully homomorphic encryption (FHE).

Concretely, I fully implemented our two main approaches and variants thereof on top of the Lattigo FHE library. This modular implementation closely follows the Lattigo interface, allowing existing users of the library to effortlessly add integrity-protection to their existing FHE pipelines, with minimal changes.

While there have been some attempts to add integrity guarantees to FHE, very few of theses approaches have actually been implemented in practice, and these approaches do not support state-of-the-art FHE schemes. This implementation is the first to provide provable integrity guarantees for state-of-the-art FHE schemes, at a very moderate overhead (×1.5 - ×50).

Work done with Sylvain Chatel, Apostolos Pyrgelis, and Jean-Pierre Hubaux at EPFL’s Lab for Data Security (LDS) as part of the Summer@EPFL summer school, followed by a student research assistant postion, both in EPFL’s Laboratory for Data Security (LDS)