Both sides of the prior authorization transaction have automated. Payers use models to triage and decide; provider organizations use models to assemble and submit. The measurable result is speed: median turnaround at several large plans has fallen from days to under an hour for routine requests.

The qualitative result is a growing category of denial that clinicians describe as unfalsifiable. Denials increasingly cite a synthesis of guidelines rather than a specific criterion, which makes them hard to rebut with a specific fact. Appeal volume is up, and appeal success is up too, which suggests a meaningful share of the fast denials were wrong.

Regulators in several states have started asking plans to disclose whether a human reviewed a denial before it was issued, and at what level of detail. Early responses have been vague enough that at least one state insurance department has issued follow-up requests.

The provider-side automation raises its own question that few are asking loudly: if both the request and the denial are machine-generated, and the appeal is machine-assembled, it is not obvious which step in the process is still doing the thing prior authorization was invented to do.