본문 바로가기
  • Home

A Zero-Relationship Lattice Algorithm for Preventing AI-Based Secret-Key Recovery Attacks on Lattice Cryptography

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2026, 31(1), pp.121~132
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : November 27, 2025
  • Accepted : December 31, 2025
  • Published : January 30, 2026

Jian Woo 1 Seungmin Lee 1 Jungwoong Jeun 1 Sejong Lee 1

1영남대학교

Accredited

ABSTRACT

Recently, attack techniques that exploit AI-based pattern analysis, dimensionality reduction, and statistical imbalance have revealed that lattice-based cryptography is not sufficiently secure when it relies solely on mathematical hardness. This paper proposes a zero-relation lattice algorithm designed to thwart AI-driven secret key recovery attacks on lattice-based cryptosystems. The proposed method applies a Maclaurin-series-based keyed polynomial obfuscation function to both the secret key and the plaintext encoding in the encryption process, transforming the linear traces remaining in the ciphertext into nonlinear distortions. Since the obfuscation coefficients are randomly derived from a secret parameter  for each key, the statistical structure observed by an adversary concentrates on an obfuscated hallucination key ′     rather than on the original secret key s. As a result, Transformer-based learning attacks such as SALSA converge to ′ instead of , and without knowledge of the secret parameter  it becomes computationally difficult to efficiently infer  from ′ The plaintext is further protected by an additional masking mechanism that depends on , so that message confidentiality is preserved even when both the ciphertext and the obfuscated secret key are exposed. Experimental results show that the proposed scheme reduces the success rate of AI-based pattern-analysis key recovery attacks by more than 70% and degrades the effectiveness of attacks that exploit dimensionality reduction and statistical imbalance, thereby achieving improved security compared with conventional lattice-based cryptosystems.

Citation status

* References for papers published after 2024 are currently being built.