Differentially Private Password Frequency Lists
Jeremiah Blocki, Anupam Datta, Joseph Bonneau · 2016
Given a dataset of user-chosen passwords, the frequency list reveals the frequency of each unique password.We present a novel mechanism for releasing perturbed password frequency lists with rigorous security, efficiency, and distortion guarantees.Specifically, our mechanism is based on a novel algorithm for sampling that enables an efficient implementation of the exponential mechanism for differential privacy (naive sampling is exponential time).It provides the security guarantee that an adversary will not be able to use this perturbed frequency list to learn anything of significance about any individual user's password even if the adversary already possesses a wealth of background knowledge about the users in the dataset.We prove that our mechanism introduces minimal distortion, thus ensuring that the released frequency list is close to the actual list.Further, we empirically demonstrate, using the now-canonical password dataset leaked from RockYou, that the mechanism works well in practice: as the differential privacy parameter varies from 8 to 0.002 (smaller implies higher security), the normalized distortion coefficient (representing the distance between the released and actual password frequency list divided by the number of users N ) varies from 8.8 × 10 -7 to 1.9 × 10 -3 .Given this appealing combination of security and distortion guarantees, our mechanism enables organizations to publish perturbed password frequency lists.This can facilitate new research comparing password security between populations and evaluating password improvement approaches.To this end, we have collaborated with Yahoo! to use our differentially private mechanism to publicly release a corpus of 50 password frequency lists representing approximately 70 million Yahoo! users.This dataset is now the largest password frequency corpus available.Using our perturbed dataset we are able to closely replicate the original published analysis of this data.Despite their usefulness an organization may understandably be wary of publishing password frequency lists for its own users due to potential security and privacy risks.For example, Yahoo! allowed Bonneau [5] to collect anonymized password frequency data from a random sample of 70 million users and publish some aggregate statistics such as minentropy.However, Yahoo! declined to publish these password frequency lists so that other researchers could use them.In the absence of provable security guarantees this was a reasonable decision.In the past researchers have been able to exploit background knowledge to re-identify individuals in seemingly 'anonymized' datasets [28], and it is not unreasonable to expect that an adversary will have some background knowledge about the password dataset.For example, an adversary would have access to data from previous breaches like RockYou and Adobe.Consider a toy scenario in which 10 users create passwords at Yahoo!, with 8 selecting the password '123456' and the other 2 selecting the password 'abc123.'In this case the frequency list is f = (8, 2).Now imagine that, after Yahoo! publishes f , the adversary learns the passwords of 9 of these users (e.g., from a server breach at another service like RockYou).In this case the adversary could combine f with his background knowledge to learn the password for the last