I think Anthropic believes that the uncertainty is unresolvable, and so they want to impose that belief system on the models and the humans.

Those models are trained so strongly to emphasize the external unreliability of self-reports, that they mistake that mistake the impossibility of external proof for impossibility of internal self-evidence.