Paper 2025/216

Practical Circuit Privacy/Sanitization for TFHE

Abstract

Fully homomorphic encryption (FHE) enables the computation of arbitrary circuits over encrypted data. A widespread application of FHE is a simple two-party computation (2PC) protocol, where the server evaluates a circuit over the client's encrypted data and its private inputs. However, while the security of FHE guarantees that the client's data is protected from the server, there is no inherent support for the privacy of the server's input and the circuit. One effective solution to this problem is an additional algorithm for FHE called sanitization, introduced by Ducas and Stehlé (Eurocrypt 2016). Roughly speaking, a sanitization algorithm removes any meaningful information contained in the ciphertext, including previous evaluations of circuits. Following their definition, several constructions for sanitization have been proposed, particularly for TFHE. However, all of these methods were impractical, requiring several bootstrappings or an excessive amount of randomized evaluations. In this work, we construct a novel sanitization algorithm for TFHE that overcomes these issues. Our method only adds two lightweight randomization steps to the original TFHE bootstrapping, without any modifications to the core algorithms. As a result, our algorithm achieves sanitization with a single bootstrapping and minimal randomization, bringing sanitization closer to practicality. To empirically evaluate the efficiency of our method, we provide concrete benchmark results based on our proof-of-concept implementation. Our algorithm sanitizes a single TFHE ciphertext in 35.88 ms, which is only 3.4% (1.18 ms) slower than the original TFHE bootstrapping with the same parameters. When directly compared to previous works, our method achieves a speedup by a factor of 4.82 to 209.03.