Memorization occurs where machine learning models contain (parts of) their training data. Memorized data points can encode sensitive information about people and persist even when the training data itself have been deleted, continuing to influence the model’s predictions. This chapter explores ethical issues related to memorization as grounds for moral complaints against accessing or utilizing memorized data points. It also discusses corresponding remediating obligations, especially machine unlearning as a means to make models forget memorized data, but also others, such as dismissing model predictions as inadmissible evidence. The chapter suggests that there can be a duty to make models unlearn, similar to how people can be obligated to forget information, and that, like forgetting as a cognitive mechanism, machine unlearning can facilitate numerous goods even in the absence of such duties.