Outputs that mislead a recipient, whether or not the model has an internal mechanism designed to deceive.