AUC measures the power of discrimination and it ignores the predicted probability values and the goodness-of-fit of the model. It is possible that a well fitted model has poor discrimination, if probabilities for positives are only moderately higher than those for negatives. It summarizes the test performance over regions of the ROC space in which one would rarely operate. For example, general ml tasks would rarely operate in the extreme left hand side (high false negative) of the curve and extreme right hand side (high false positive). For example in medical diagnostics, false negatives are more expensive than false positives. It does not give information about the spatial distribution of model errors.