Any AI system deployed to make decisions that affect human outcomes for which the human doesn’t have an opt out, should be subject to independent audits of the training data and model by credible organisations