MOZI: Threshold FHE at Scale: Efficient and Resilient, for Large-Scale MPC

"Preview Talk" (by team MOZI) @ TCPT3, in reply to the NIST Threshold Call

Abstract: In this presentation, I will introduce MOZI and discuss how to make threshold fully homomorphic encryption (ThFHE) scalable for hundreds or thousands of participants. ThFHE enables computation on encrypted data while distributing decryption authority among multiple key holders. Scaling this capability requires careful control of computation, communication, and storage complexity. I will explain MOZI’s modular design, which uses approximate secret sharing (ApproxSS) to separate threshold decryption from the underlying encryption scheme. This design supports two complementary systems: MOZI-Fang uses lattice-based symmetric encryption and targets post-quantum security, while MOZI-Yuan combines lattice techniques with elliptic-curve operations to explore lower recovery costs under quantum-vulnerable assumptions. These goals include guaranteed output delivery with an available semi-honest combiner and sufficient honest participation, and security with abort against a malicious combiner. Finally, I will outline our evaluation plan, covering network costs, optional preprocessing, and authenticated Beaver-triple generation for secure multi-party computation. The talk will highlight the design choices and practical tradeoffs involved in scaling threshold encryption for collaborative applications.

Joint work: Man Ho Au, Yijia Chang, Rongmao Chen, Xinyi Huang, Xianhui Lu, Moti Yung.

Suggested readings

• Preview Writeup (PDF): MOZI: Threshold FHE at Scale: Efficient and Resilient, for Large-Scale MPC
• Arbitrary-Threshold Fully Homomorphic Encryption with Lower Complexity (ia.cr/2025/084)
• Resolving the Efficiency-Utility Dilemma of Threshold Linearly Homomorphic Encryption via Message-Space Adapter (ia.cr/2025/948)