Hardening Confidential Federated Compute against Side-channel Attacks

TL;DR

This paper identifies side-channels in a confidential federated compute platform that threaten privacy guarantees and demonstrates how differential privacy can mitigate some of these vulnerabilities, including implementation in an open-source library.

Contribution

The work uncovers specific side-channels in federated compute platforms and shows how differential privacy can mitigate these risks, with practical implementation.

Findings

DP mitigates two identified side-channels

One side-channel has been implemented in open-source code

Highlights importance of security analysis in federated compute

Abstract

In this work, we identify a set of side-channels in our Confidential Federated Compute platform that a hypothetical insider could exploit to circumvent differential privacy (DP) guarantees. We show how DP can mitigate two of the side-channels, one of which has been implemented in our open-source library.