One Error to Rule Them All: Can a Single Bit-Flip Disrupt Fully Homomorphic Encryption?

Fully Homomorphic Encryption (FHE) enables secure computation on encrypted data without decryption, representing a major advance in privacy-preserving technologies. This capability is especially valuable for safety-critical domains such as healthcare, finance, and government, where sensitive data must remain protected while enabling third-party computations. Despite its importance, systematic studies of FHE’s resilience—a critical concern for high-reliability applications—remain largely unexplored.

In this work, we investigate the resilience of CKKS (Cheon-Kim-Kim-Song), a widely used FHE scheme that supports approximate arithmetic for privacy-preserving machine learning and scientific simulations.We provide a comprehensive evaluation of CKKS under single-bit and multi-bit hardware errors, focusing on how Residue Number System (RNS) and Number Theoretic Transform (NTT) optimizations affect (worsen) error propagation. Across over 175 million error-injection experiments spanning 500 parameter configurations, we identify error-resilience patterns that are general, data-independent, and consistent across FHE pipeline variations. Our findings offer a foundation for further work in designing fault-tolerant cryptographic systems.