Skip to content
ContestabilityA research
publication
HomeWorkshop response

Making AI-assisted insurance claims open to meaningful challenge

A California Department of Insurance workshop response asks what claimants need to investigate AI-assisted denials, using a fictional roof claim and six proposed safeguards.

Author
Kerry Thornhill
Status
Workshop response · Submitted November 15, 2026
Source date
Web version
2026-11-15 · v1

Kerry Thornhill

MS in Interdisciplinary Studies · In progress · University of North Texas

Anuradha and Vikas Sinha College of Artificial Intelligence and Advanced Analytics

A tree fallen across the roof of a house that stands on the floor of a mainframe computer room

Workshop response · November 2026

Making AI-assisted insurance claims open to meaningful challenge

Kerry Thornhill

An insurer can record a roof’s age correctly while leaving the cause of its damage in dispute. If an AI-assisted denial rests on the wrong image, an omitted report, or an unsupported inference, what must the claimant be able to investigate?

I submitted written comments to the California Department of Insurance for its December 15 discussion workshop, Exploring the Responsible Use of Artificial Intelligence in Insurance (REG-2026-00007). The response applies the Necessary Informational Conditions (NIC) framework to explanations, evidence handling, human review, and vendor accountability in insurance claims.

A correct input can support a disputed conclusion

The response develops a fictional property claim. A policyholder submits photographs and a contractor’s report identifying fresh storm damage. An automated assessment supports denial under a stipulated wear exclusion, and a human adjuster signs the letter. The roof’s age is accurate, but the assessment used an earlier image and did not incorporate the contractor’s report.

Claimant: “You have confirmed the roof’s age. Which images and report did you consider, and what supports your conclusion that this damage came solely from wear?”

Illustrative inadequate response: “The age field is accurate, and an adjuster reviewed the result.”

The claimant already has grounds to object. The further question is whether there is a feasible way to investigate the insurer’s answer. Confirming age or a human signature leaves the causal objection unanswered. The example keeps the outcome open: substantive review could support reversal, a reasoned decision to maintain the denial, or further investigation.

Six proposed safeguards

  1. Make material grounds investigable: connect the evidence, relevant policy provision, and determination in this claim.
  2. Make evidence use traceable: distinguish receiving a report from considering it, and identify relevant dates and versions.
  3. Test the response to the objection: give a competent reviewer the access and authority to address the disputed inference.
  4. Preserve insurer accountability across vendors: ensure that the insurer can investigate defects and provide a clear institutional contact.
  5. Provide protected but usable access: test whether tailored explanations or intermediaries enable meaningful inquiry.
  6. Record disposition and follow-through separately: distinguish the answer to an objection, any correction, and its implementation.

These are implementation and research proposals. The response separately identifies existing California requirements for first-party denial explanations and Department-review information. It does not transfer employment-specific rules into insurance or claim that every proposed document is legally required.

Test capability, including successful alternatives

The proposed evaluation compares different feasible routes through controlled cases. If a competent reviewer enables the same investigation without an evidence inventory, that inventory has not been shown necessary. Conversely, providing an inventory is insufficient when the review process does not address the objection. Investigative capability, institutional response, and implemented consequences should be measured separately.

Read the submitted response (PDF, 4 pages)

The response was sent by email on November 15; a Department acknowledgment has not yet been recorded. This is a discussion-workshop submission. Any later formal rulemaking may require separate comments. See the official invitation and discussion topics.

The framework is developed in Paper A, From Notice to Contestable Grounds. Related applications include the California workplace AI example and the submitted ICO response. The insurance case is fictional and the proposals remain open to empirical scrutiny.

Return to Articles and Activities