Model developers may need to delete data from ML systems for several reasons, including privacy regulations that stipulate a right to be forgotten and/or to enable debugging. Unfortunately, retraining models where the deleted data is omitted is expensive and time-consuming. In this paper, Izzo et al. describe an approximate deletion method for linear and logistic models with lower computational costs and higher accuracy. In addition, they propose a feature injection test that evaluates how well a deletion model removes the model’s knowledge of sensitive attributes of the deleted data points.