When Human Judgment Becomes Difficult to Verify
Iowa's House File 2635 became effective July 1. For prior authorization requests beginning January 1, 2027, it allows artificial intelligence to conduct an initial review while prohibiting AI from serving as the sole basis for denying, delaying, or downgrading a request involving medical necessity. A qualified reviewer or clinical peer must make a denial or downgrade, depending on the type of provider requesting authorization, and sign a written statement explaining the reasons and criteria behind it. The documentation requirement is familiar. The judgment requirement is different. Rather than regulating only a document or workflow, HF 2635 also regulates the role AI can play before an adverse determination is reached. That raises a practical question once the review is complete: what evidence would demonstrate that AI was not the sole basis for the decision?
HF 2635 Regulates the Decision Before It Regulates the Record
Most administrative requirements govern something tangible that can be directly observed. A notice is sent. A form is completed. A policy is followed. HF 2635 reaches further upstream. The requirement is not simply that a reviewer sign an adverse determination. The appropriate reviewer must make the decision, and AI cannot serve as its sole basis.
The distinction exists because the decision-making process takes place before anything is documented. Whether a clinician substantively evaluated the request or relied primarily on an existing AI recommendation is part of how the decision was made, not part of the record created afterward. The restriction applies before the documentation ever exists.
Documentation Shows the Rationale, Not the Judgment
Once the review is complete, the remaining evidence is administrative. The record identifies the reviewer, cites the specific reasons and criteria behind the determination, and includes the reviewer's signature and professional qualifications. Those records support oversight, appeals, and compliance because they establish what decision was documented and who accepted responsibility for it.
The reasoning process that produced the decision has already ended. Two reviewers who arrive at the same determination through different paths may leave behind similar written explanations. One may substantively evaluate the clinical record and use AI as one source of assistance. Another may rely heavily on an AI recommendation while adding limited human input before signing the same rationale. Looking only at the completed record, those paths may be difficult to distinguish.
Organizations Are Already Defining the Process Themselves
That question has already begun shaping implementation. Guidance from Paul Drey, a Des Moines healthcare attorney at Brick Gentry and Emily Reiners encourages organizations to identify where artificial intelligence is currently used, define where human oversight must occur, document those workflows internally, and ensure governance reflects how these tools are actually used in practice.
The guidance extends beyond documentation itself. HF 2635 establishes who must make an adverse determination and prevents AI from serving as its sole basis. Organizations are now being encouraged to define where human review begins, how it will be documented, and how that requirement will operate in practice.
The Prior Auth Report identifies the prior authorization developments that actually matter each month and explains what they mean operationally, for the people making decisions and the people doing the work.
If you want that analysis delivered directly to your inbox each month, the newsletter signup is below.