In this paper, we describe a brand new key exchange protocol based on a semidirect product of (semi)groups (more specifically, on extension of a (semi)group by automorphisms), and then focus on practical instances of this general idea. Our protocol can be based on any group, in particular on any non-commutative group. One of its special cases is the standard Diffie-Hellman protocol, which is based on a cyclic group. However, when our protocol is used with a non-commutative (semi)group, it acquires several useful features that make it compare favorably to the Diffie-Hellman protocol. Here we also suggest a particular non-commutative semigroup (of matrices) as the platform and show that security of the relevant protocol is based on a quite different assumption compared to that of the standard Diffie-Hellman protocol.

Abstract. We offer a public key exchange protocol in the spirit of Diffie-Hellman, but we use (small) matrices over a group ring of a (small) symmetric group as the platform. This "nested structure" of the platform makes computation very efficient for legitimate parties. We discuss security of this scheme by addressing the Decision Diffie-Hellman (DDH) and Computational Diffie-Hellman (CDH) problems for our platform.

The Anshel-Anshel-Goldfeld (AAG) key-exchange protocol was implemented and studied with the braid groups as its underlying platform. The length-based attack, introduced by Hughes and Tannenbaum, has been used to cryptanalyze the AAG protocol in this setting. Eick and Kahrobaei suggest to use the polycyclic groups as a possible platform for the AAG protocol. In this paper, we apply several known variants of the length-based attack against the AAG protocol with the polycyclic group as the underlying platform. The experimental results show that, in these groups, the implemented variants of the length-based attack are unsuccessful in the case of polycyclic groups having high Hirsch length. This suggests that the length-based attack is insu cient to cryptanalyze the AAG protocol when implemented over this type of polycyclic groups. This implies that polycyclic groups could be a potential platform for some cryptosystems based on conjugacy search problem, such as non-commutative Di e-Hellman, El Gamal and Cramer-Shoup key-exchange protocols. Moreover, we compare for the rst time the success rates of the di erent variants of the length-based attack. These experiments show that, in these groups, the memory length-based attack introduced by Garber, Kaplan, Teicher, Tsaban and Vishne does better than the other variants proposed thus far in this context.