Abstract or Summary:
We propose to strengthen user-selected passwords against statistical-guessing attacks by allowing users of Internet-scale systems to choose any password they want--so long as it's not already too popular with other users. We create an oracle to identify undesirably popular passwords using an existing data structure known as a count-min sketch, which we populate with existing users' passwords and update with each new user password. Unlike most applications of probabilistic data structures, which seek to achieve only a maximum acceptable rate false-positives, we set a minimum acceptable false-positive rate to confound attackers who might query the oracle or even obtain a copy of it.