Recently, by an interesting confluence, multivariate schemes with the minus modifier have received attention as candidates for multivariate encryption. Among these candidates is the twenty year old HFE– scheme originally envisioned as a possible candidate for both encryption and digital signatures, depending on the number of public nist-equations removed. HFE has received a great deal of attention and a variety of cryptanalyses over the years; however, HFE– has escaped these assaults. The direct algebraic attack that broke HFE Challenge I is provably more complex on HFE– , and even after two decades HFE Challenge II is daunting, though not achieving a security level we may find acceptable today. The minors modeling approach to the Kipnis-Shamir (KS) attack is very efficient for HFE, but fails when the number of nist-equations removed is greater than one. Thus it seems reasonable to use HFE– for encryption with two nist-equations removed. This strategy may not be quite secure, however, as our new approach shows. We derive a new key recovery attack still based on the minors modeling approach that succeeds for all parameters of HFE–. The attack is polynomial in the degree of the extension, though of higher degree than the original minors modeling KS-attack. As an example, the complexity of key recovery for HFE– (q=31,n=36,D=1922,a=2) is 252 . Even more convincingly, the complexity of key recovery for HFE Challenge-2, an HFE– (16,36,4352,4) scheme, is feasible, costing around 267 operations. Thus, the parameter choices for HFE– for both digital signatures and, particularly, for encryption must be re-examined.
June 26-28, 2017
PQCrypto 2017: The Eighth International Conference on Post-Quantum Cryptography
Multivariate cryptography, HFE, Encryption, MinRank, Q-rank