A new preprint on arXiv (2610.24477) studies the problem of differentially private black-box evaluation of arbitrary high-dimensional functions. The setting assumes an analyst has a dataset S and query access to a function f, but the function's sensitivity may be unknown or unbounded. This makes standard differential privacy mechanisms difficult to apply directly.

The abstract is truncated, so the full proposed approach is not described in the available text. However, the stated goal is clear: to privately evaluate such functions despite the lack of sensitivity bounds. This is a meaningful step toward practical privacy for complex, high-dimensional computations where worst-case sensitivity cannot be easily computed.

As only one source is provided, there are no conflicting accounts to compare. The significance lies in addressing a known gap in differential privacy—handling functions with unknown or unbounded sensitivity in a black-box setting. Further details would require the full paper.