Differential privacy libraries rely on sampling from exact noise distributions to provide formal guarantees. A new arXiv preprint reports the discovery of systematic periodic artifacts in OpenDP's discrete Laplace sampler, a key component for generating such noise.
The artifacts were found in the sampler's output distribution. Because the privacy guarantees depend on precise sampling, even subtle deviations are potentially significant. The authors detail how the artifacts were detected and describe a resolution for the sampler.
The paper (arXiv:2610.01907) highlights the importance of auditing the implementations of privacy primitives, not just their theoretical designs, to ensure that deployed systems deliver the privacy they promise.