Noting that training data and the model's training process can carry bias, including latent bias not visible in the data, take reasonable measures to manage data quality; and, since bias cannot be wholly removed from data and training, ensure that models are trained on representative datasets and that the AI system is checked for unfair bias, meaning bias that brings a particular individual or group a disadvantage that cannot be rationally explained.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.