When do latent class models overstate accuracy for diagnostic and other classifiers in the absence of a gold standard?

When do latent class models overstate accuracy for diagnostic and other classifiers in the absence of a gold standard?