Breaking Differential Privacy with a Stopwatch: Attacks and Formal Defenses

Differential privacy (DP) provides a rigorous foundation for protecting individual privacy in statistical data analysis and is now a cornerstone of privacy-preserving systems. Intuitively, DP ensures that an algorithm’s output will not change in a meaningful way if any single individual’s data is added or removed from the input. However, real-world implementations of DP often leak exploitable information through observable behaviors beyond the algorithm’s output, which are not considered by standard proofs that establish DP guarantees. One particularly important example is timing side channels, which arise when an adversary can observe a DP algorithm’s runtime. 

This talk is based on joint work appearing in CCS’24 (Distinguished Artifact Award, joint with Salil Vadhan), CCS’25 (joint work with Nicolas Berrios and James Mickens), and TCC’25 (joint work with Salil Vadhan).  Please see abstract for more details