Invited talk on open problems in machine unlearning for AI safety at Ploutos.