Abstract
Many organisations enforce policies on the length and formation of passwords to encourage selection of strong passwords and protect their multi-user systems. For Two-Server Password Authenticated Key Exchange (2PAKE) and Two-Server Password Authenticated Secret Sharing (2PASS) protocols, where the password chosen by the client is secretly shared between the two servers, the initial remote registration of policy-compliant passwords represents a major problem because none of the servers is supposed to know the password in clear. We solve this problem by introducing Two-Server Blind Password Registration (2BPR) protocols that can be executed between a client and the two servers as part of the remote registration procedure. 2BPR protocols guarantee that secret shares sent to the servers belong to a password that matches their combined password policy and that the plain password remains hidden from any attacker that is in control of at most one server. We propose a security model for 2BPR protocols capturing the requirements of policy compliance for client passwords and their blindness against the servers. Our model extends the adversarial setting of 2PAKE/2PASS protocols to the registration phase and hence closes the gap in the formal treatment of such protocols. We construct an efficient 2BPR protocol for ASCII-based password policies, prove its security in the standard model, give a proof of concept implementation, and discuss its performance.