Generating 56-bit passwords using Markov Models (and Charles Dickens)

TL;DR

This paper presents a method for generating 56-bit passwords using Markov models trained on Dickens' text, ensuring uniform probability and a balance between memorability and security.

Contribution

It introduces a bijective mapping between random bitstrings and passwords generated from a Markov model based on Dickens' text, ensuring equal probability for all passwords.

Findings

Passwords are memorable and secure, balancing human usability and cryptographic strength.

The system produces passwords like 'The cusay is wither?" t' with 56-bit entropy.

The method guarantees uniform distribution over generated passwords.

Abstract

We describe a password generation scheme based on Markov models built from English text (specifically, Charles Dickens' *A Tale Of Two Cities*). We show a (linear-running-time) bijection between random bitstrings of any desired length and generated text, ensuring that all passwords are generated with equal probability. We observe that the generated passwords appear to strike a reasonable balance between memorability and security. Using the system, we get 56-bit passwords like 'The cusay is wither?" t', rather than passwords like 'tQ$%Xc4Ef'.