All posts
Insurance

AI in Underwriting and Claims in 2026: The NAIC Bulletin, Colorado, and the EU AI Act's New Dates

By

October 26, 2026

7 min read

Twenty-nine jurisdictions had adopted the NAIC's AI model bulletin by October 2026, Colorado replaced its general AI law in May 2026 while keeping its insurance-specific rules, and the EU moved high-risk obligations, including life and health insurance pricing, to December 2, 2027. What insurers using AI need to be able to show, and how to build it.

AI Governance
Insurance
NAIC
EU AI Act
Underwriting

Regulators have mostly not banned anything. They expect insurers to know which AI systems they use, to govern them, and to be able to show the results are not unfairly discriminatory. That is an engineering problem as much as a legal one.

The rules, as of October 2026

Rule

Status

What it expects

NAIC Model Bulletin on the Use of AI Systems by Insurers

Adopted by the NAIC December 4, 2023; by 29 jurisdictions as of October 8, 2026, with California, Colorado, New York and Texas using their own insurance-specific rules or guidance

A written AI program: governance, risk management, internal controls, oversight of third-party models and data

Colorado 3 CCR 702-10

Effective November 13, 2023, amended effective October 15, 2025

Governance and risk management for external consumer data, algorithms and predictive models, with testing for unfair discrimination

Colorado SB26-189

Signed May 14, 2026; repeals and re-enacts the 2024 Colorado AI Act; effective January 1, 2027

Notices to consumers, developer documentation for deployers, and a right to request human review after adverse automated decisions

EU AI Act, as amended by Regulation (EU) 2026/1744

Omnibus in force July 27, 2026; Annex III high-risk obligations apply from December 2, 2027

Risk assessment and pricing for natural persons in life and health insurance is high-risk: risk management, data governance, logging, human oversight, documentation

Want a second opinion on this for your system?

Book a free 30-minute call. You'll leave knowing what we'd do first and roughly what it costs, whether or not you hire us.

What you need to be able to show

  1. An inventory of AI systems and predictive models: purpose, owner, data sources, decisions affected, vendor if any.

  2. Documentation for each: how it was built or selected, what it was tested for, known limits.

  3. Testing for unfair discrimination, repeated when the model or data changes.

  4. Human review for adverse decisions where rules require it, and an explanation path for consumers.

  5. Logs of inputs, outputs and model versions for decisions, kept long enough to answer a regulator or a complaint.

  6. Oversight of vendors: contract rights to information and audits, not only a questionnaire.

Building it on AWS

  • Model registry and documentation: SageMaker Model Registry and model cards for trained models; the same inventory for LLM features, recording the Bedrock model ID and prompt version.

  • Decision logs: every scored application or claim written with inputs, output, model version and reviewer to an immutable store such as S3 with Object Lock.

  • Evaluation: scheduled bias and accuracy tests on held-out data, and Bedrock evaluations for generative features, with results stored next to the model version.

  • Human review: Amazon A2I or your own work queue for low-confidence or adverse outcomes.

hoolco Monthly Review

Once a month: what changed in tech that month and what it means for you, from cloud and AI to security and the tools you build with. Every item links to its source. No sales pitch, unsubscribe in one click.

As of October 26, 2026. Provided as is, for general information, and accurate to the best of our knowledge on the date shown. Not legal, compliance, tax, financial or other professional advice. Prices, limits and vendor services change; check current vendor documentation before relying on anything here. Terms of use.