The Measure of Deception: An Analysis of Data Forging in Machine Unlearning

ArXi:2509.05865v2 Announce Type: replace Motivated by privacy regulations and the need to mitigate the effects of harmful data, machine unlearning seeks to modify trained models so that they effectively ``forget'' designated data. A key challenge in verifying unlearning is \emph{forging} -- adversarially crafting data that mimics the gradient of a target point, thereby creating the appearance of unlearning without actually removing information.