A Bio-Inspired Authenticated Key Exchange Binding AlphaFold2 Protein Geometry to Ephemeral Elliptic-Curve Diffie-Hellman.

Authenticated key exchange has been built for decades on a small number of hard mathematical problems, and the elliptic curve discrete logarithm is one of them. The structural complexity of biological macromolecules, however, has rarely been used as a source of secret material in such protocols. In this paper, a protein-derived authenticated key exchange protocol, called PDKE-2, is proposed to address this gap. The protocol keeps elliptic curve Diffie-Hellman at its core and adds two ingredients from structural biology. Firstly, a per-session base point G* is generated by applying RFC 9380 hash-to-curve to a session mapping table. Secondly, a long-term secret shared between the two parties is derived from the SHA-256 digest of an AlphaFold2 inter-residue distance matrix. In contrast to an earlier version of this design, the two parties run ephemeral Diffie-Hellman in each session and the session key is never sent over the channel; only transcript-based confirmation tags are exchanged. Thus, the protocol provides forward secrecy and avoids a confidentiality weakness that existed in the earlier version. The security of PDKE-2 is analysed in the Real-or-Random model, where session key secrecy, forward secrecy and mutual authentication are proved under the Gap Diffie-Hellman, PRF-HKDF and EUF-CMA-HMAC assumptions. In addition, the protocol is modelled in ProVerif against a Dolev-Yao attacker, and the secrecy of the session key, the secrecy of the shared secret and the injective agreement in both directions are all confirmed. A Python reference implementation is also provided. In the implementation, key agreement and mutual authentication succeed in every session, and the complete handshake takes about 15 ms in the reference code. It should be noted that the protein layer does not increase the elliptic curve security level; it works as an authentication secret whose strength depends on the difficulty of guessing the protein sequence when this sequence is kept private.