CATCH researchers PI Olya Ohrimenko, PhD student Jiankai Jin and PI Ben Rubinstein have presented their work on systems implementation flaws in differential privacy, last week at the prestigious 2022 IEEE Symposium on Security & Privacy (also known as ‘Oakland’) in San Francisco. Demonstrating new flaws in floating-point rounding and timing side-channel attacks, found in industry-standard implementations by Google and Facebook/Meta, the work was recognized by an Honorable Mention in Google’s Bug Hunter program due to its impact on the Google Differential Privacy library. The work also resulted in a patch to Facebook/Meta’s Opacus library for privacy-preserving deep learning.