Paper 2024/203

Application-Aware Approximate Homomorphic Encryption: Configuring FHE for Practical Use

Andreea Alexandru, Duality Technologies
Ahmad Al Badawi, Duality Technologies
Daniele Micciancio, University of California, San Diego, Duality Technologies
Yuriy Polyakov, Duality Technologies
Abstract

Fully Homomorphic Encryption (FHE) is a powerful tool for performing computations on encrypted data. The Cheon-Kim-Kim-Song (CKKS) scheme, an instantiation of approximate FHE, is particularly effective for privacy-preserving machine learning applications over real and complex numbers. Although CKKS offers clear efficiency advantages, confusion persists around accurately describing applications in FHE libraries and securely instantiating the scheme for these applications, particularly after the key recovery attacks by Li and Micciancio (EUROCRYPT'21) for the setting. There is presently a gap between the application-agnostic, generic definition of , and efficient, application-specific instantiation of CKKS in software libraries, which led to recent attacks by Guo et al. (USENIX Security'24). To close this gap, we introduce the notion of application-aware homomorphic encryption (AAHE) and devise related security definitions. This model corresponds more closely to how FHE schemes are implemented and used in practice, while also identifying and addressing the potential vulnerabilities in popular libraries. We then provide an application specification language (ASL) and formulate guidelines for implementing the AAHE model to achieve security for practical applications of CKKS. We present a proof-of-concept implementation of the ASL in the OpenFHE library showing how the attacks by Guo et al. can be countered. Moreover, we show that our new model and ASL can be used for the secure and efficient instantiation of exact FHE schemes and to counter the recent attacks by Cheon et al. (CCS'24) and Checri et al. (CRYPTO'24).

Metadata
Available format(s)
PDF
Category
Public-key cryptography
Publication info
Preprint.
Keywords
application-aware homomorphic encryptionapproximate FHECKKS
Contact author(s)
aalexandru @ dualitytech com
aalbadawi @ dualitytech com
daniele @ cs ucsd edu
ypolyakov @ dualitytech com
History
2025-01-28: last of 3 revisions
2024-02-09: received
See all versions
Short URL
https://ia.cr/2024/203
License
Creative Commons Attribution
CC BY

BibTeX

@misc{cryptoeprint:2024/203,
      author = {Andreea Alexandru and Ahmad Al Badawi and Daniele Micciancio and Yuriy Polyakov},
      title = {Application-Aware Approximate Homomorphic Encryption: Configuring {FHE} for Practical Use},
      howpublished = {Cryptology {ePrint} Archive, Paper 2024/203},
      year = {2024},
      url = {https://eprint.iacr.org/2024/203}
}
Note: In order to protect the privacy of readers, eprint.iacr.org does not use cookies or embedded third party content.