CHORUS is Porous: Attacking Implementations of Differential Privacy

Abstract

In this work I explore the vulnerability of CHORUS and FLEX, using a side- channel attack. CHORUS and FLEX are differentially private querying mecha- nisms jointly worked on by Uber and UC Berkeley. They aim to provide noisy results to an analyst’s queries. I propose and implement an algorithm to use the exploit, and analyze its efficacy. Finally, I implement the proposed defense and verify that it protects against this exploit.