A large language model that is trained using AI outputs can inherit undesirable behaviours, even if they are not directly referenced in the training data. Work this year has shown that AI models that ...