What's in a name? Evaluating statistical attacks on personal knowledge questions

We study the efficiency of statistical attacks on human authentication systems relying on personal knowledge questions. We adapt techniques from guessing theory to measure security against a trawling attacker attempting to compromise a large number of strangers' accounts. We then examine a diverse corpus of real-world statistical distributions for likely answer categories such as the names of people, pets, and places and find that personal knowledge questions are significantly less secure than graphical or textual passwords. We also demonstrate that statistics can be used to increase security by proactively shaping the answer distribution to lower the prevalence of common responses.